JP3989787B2 - Compression encoding apparatus, compression encoding method and program - Google Patents

Description

【００６４】
上式（１）中、ｚは、ビット切り捨て点（bit truncation point）；ｏｙ_i ^K[i,j]［ｊ］は、Ｋ［ｉ，ｊ］番目のビットプレーンで逆量子化されたコードブロックのｊ番目のサンプル値（係数値）；ｙ_i［ｊ］は、当該コードブロックのｊ番目のサンプル値（係数値）；Ｇ_b[i]は、サブバンドｂ［ｉ］に対応する合成フィルタ係数のノルムの二乗であって、当該サブバンドｂに依存する歪モデルの重み係数を示している。尚、説明の便宜上、上式（１）に示した記号の表記法は、参考文献Ａでのそれとは若干異なる。
【００６５】
レート・歪み最適化では、この歪測度Ｄ_i ^(z)のサブバンドｂ［ｉ］における総和量を最小にするような最適化処理が行われる。サブバンドｂの重み係数Ｇ_bは、画像の歪みを低減させるための重み付けを表している。
【００６６】
サブバンドｂの重み係数Ｇ_bは、次式（２）に従って算出される。
【００６７】
【数２】

【００６８】
ここで、上式（２）中、ｓ_b［ｎ］は、サブバンドｂの１次元合成フィルタ係数を示している。また、記号||x||は、ベクトルxに関するノルムを示す。
【００６９】
参考文献Ａに記載される数式（４．３９）と（４．４０）によれば、分解レベル１における低域成分Ｌ１の１次元合成フィルタ係数ｓ_L[1]［ｎ］と、同分解レベルにおける高域成分Ｈ１の１次元合成フィルタ係数ｓ_H[1]［ｎ］とは、次式（３）に従って算出される。
【００７０】
【数３】

【００７１】
ここで、上式（３）中、ｇ₀［ｎ］は、画像信号を帯域分割する順変換フィルタのローパス・フィルタ係数、ｇ₁［ｎ］は、そのハイパス・フィルタ係数をそれぞれ示している。
【００７２】
また、分解レベルｄ（ｄ＝１，２，…，Ｄ）における低域成分Ｌｄの１次元合成フィルタ係数ｓ_L[d]［ｎ］と、同分解レベルにおける高域成分Ｈｄの１次元合成フィルタ係数ｓ_H[d]［ｎ］とは、次式（４）に従って算出される。
【００７３】
【数４】

【００７４】
そして、分解レベルｄにおける低域成分Ｌｄの１次元合成フィルタ係数のノルムの二乗は、次式（５）に従って算出される。
【００７５】
【数５】

【００７６】
高域成分の１次元合成フィルタ係数のノルムの二乗も、上式（５）と同様にして算出することができる。
【００７７】
次に、分解レベルｄ（ｄ＝１，２，…，Ｄ；Ｄは整数）における帯域成分ＬＬＤ，ＨＬｄ，ＬＨｄ，ＨＨｄの２次元合成フィルタ係数は、上記１次元合成フィルタ係数の積で表現することができ、帯域成分ｂの２次元の重み係数Ｇ_bも、１次元の重み係数の積で表現することができる。具体的には、２次元合成フィルタ係数と２次元の重み係数とは、次式（６）に従って算出される。
【００７８】
【数６】

【００７９】
上式（６）中、添字ＬＬ［Ｄ］はサブバンドＬＬＤを示し，ＨＬ［ｄ］，ＬＨ［ｄ］およびＨＨ［ｄ］はそれぞれサブバンドＨＬｄ，ＬＨｄおよびＨＨｄを表している。
【００８０】
重み係数Ｇ_bの平方根がノルムである。以下の表１〜表４に、２次元の重み係数Ｇ_bに関する計算結果を示す。表１に、（９，７）フィルタ（９×７タップのフィルタ）の各帯域成分のノルムの二乗の数値を、表２には、表１に対応するノルムの数値をそれぞれ示す。また、表３に、（５，３）フィルタ（５×３タップのフィルタ）の各帯域成分のノルムの二乗の数値を、表４には、表３に対応するノルムの数値をそれぞれ示す。
【００８１】
【表１】

【００８２】
【表２】

【００８３】
【表３】

【００８４】
【表４】

【００８５】
更に、分解レベル１における低域成分ＬＬ１のノルムをαで表すとき、このノルムαを用いて、各帯域成分について図５に示すような値を設定する。図５に示す２次元画像２７は、オクターブ分割方式に従って帯域分割された２次元画像１２０を示す図である。分解レベルｎ（ｎ：１以上の整数）における帯域成分ＨＨｎの設定値は「２^n-3×α」、帯域成分ＨＬｎおよびＬＨｎの設定値は「２^n-2×α」、帯域成分ＬＬｎの設定値は「２^n-1×α」にそれぞれ設定される。従って、例えば、帯域成分ＬＨ１の設定値は「２^-1×α」に設定されている。
【００８６】
上記設定値と、表２と表４に示したノルムの数値とを比較すれば、両者は概ね近似する。例えば、表２の場合（α=1.96591）、図５に示す各帯域成分の「設定値（対応する帯域成分）」は、約０．４９（ＨＨ１）、約０．９８（ＨＬ１，ＬＨ１）、約１．９６（ＨＬ２，ＬＨ２，ＨＨ３）、約３．９３（ＨＬ３，ＬＨ３）、約７．８６（ＬＬ３）となり、これら設定値は、表２に示したノルムの数値と近似していることが分かる。
【００８７】
また、図５において、帯域成分ＬＬ１のノルムをα＝２に丸め込み、各帯域成分の設定値を１ビット左シフトした値、すなわち、全ての設定値に２¹を乗算した２の巾乗値の指数は、図２に示した優先度の値に一致することが分かる。よって、第１実施例のように各帯域成分に優先度を設定することは、近似的に、レート・歪み最適化で使用するフィルタのノルム（重み係数の平方根）を各帯域成分のサンプル値（変換係数値）に乗算することに等しい。従って、本実施例の優先度は画像の歪みを低減させ得るように設定されるものである。
【００８８】
優先度設定方法（第２実施例）．
次に、優先度設定方法の第２実施例について説明する。本実施例では、各帯域成分の上記重み係数Ｇ_bの平方根であるノルムを、最も高い分解レベルにおける最低域成分ＬＬのノルムで除算した値を２の巾乗値に丸め込み、その２の巾乗値の指数の絶対値を、優先度として設定する。具体的には、最も高い分解レベルｎの最低域成分ＬＬｎのノルムをαとし、その他の帯域成分のノルムをｘとし、２の巾乗に丸め込む変数ｙに関する関数をR［ｙ］とし、変数ｙの２の巾乗２^mの指数ｍを算出する関数をｍ＝Ｉ［２^m］とし、変数ｙに関する絶対値を｜ｙ｜とするとき、優先度ｐは、ｐ＝｜Ｉ［Ｒ［ｘ／α］］｜、に従って算出される。
【００８９】
以下の表５に、上記表２に示した（９，７）フィルタのノルムを用いて算出した優先度を示す。また、図６に、表５に示した優先度を記した２次元画像２８を示す帯域分割図を示す。ここで、最も高い分解レベルは５であり、α＝３３．９２４９３、である。尚、表中の「×」は、当該帯域成分の優先度は計算されていないことを意味する。
【００９０】
【表５】

【００９１】
また、以下の表６に、上記表４に示した（５，３）フィルタのノルムを用いて算出した優先度を示す。
【００９２】
【表６】

【００９３】
上記第１実施例では、各帯域成分の変換係数を優先度のビット数だけ左シフトさせて優先度を設定していたが、本第２実施例では、各帯域成分の変換係数を優先度のビット数だけ右シフトさせる処理が実行される。但し、変換係数のビット長を拡大させるように右ビットシフト処理が実行される。図７は、図６に示す優先度のビット数だけ右シフトされた帯域成分の変換係数２９，２９，…を示す模式図である。
【００９４】
優先度設定方法（第３実施例）．
次に、人間の視覚特性を考慮した第３実施例に係る優先度設定方法について説明する。数百万画素程度の高解像度の画像について上記第２実施例で示した優先度を適用した場合、復号画像の画質は客観評価では良いが、人間の視覚評価では必ずしも良いとは限らない。そこで、本実施例の優先度設定方法は、人間の視覚特性を考慮した重み付けをされた優先度を採用するものである。これにより、高い表示画質の圧縮画像を生成することが可能になる。
【００９５】
前記参考文献ＡのChapter 16には、ＣＳＦ（human visual system Contrast Sensitivity Function）に基づいた重み付けＭＳＥ（Weighted Mean Squared Error；WMSE）が記載されている。この記載によれば、人間の視覚評価を改善するために、上式（１）を次式（７）に修正するのが望ましい。
【００９６】
【数７】

【００９７】
ここで、上式（７）中、Ｗ_b[i] ^csfは、サブバンドｂ［ｉ］の"energy weighting factor"と呼ばれており、Ｗ_b[i] ^csfの推奨数値は、「ISO/IEC JTC 1/SC 29/WG1(ITU-T SG8) N2406, "JPEG 2000 Part 1 FDIS (includes COR 1, COR 2, and DCOR3)," 4 December 2001」の文献（以下、参考文献Ｂと呼ぶ。）に記載されている。図８〜図１０に、参考文献Ｂに記載される"energy weighting factor"の数値を示す。
【００９８】
図８〜図１０中の"level"および "Lev"は分解レベルを、"Comp"は輝度成分Ｙと色差成分Ｃｂ, Ｃｒをそれぞれ示しており、"Viewing distance（視距離）"が1000，1700, 2000, 3000, 4000の例が示されている。また、"Viewing distance 1000", "Viewing distance 1700", "Viewing distance 2000", "Viewing distance 3000", "Viewing distance 4000"は、それぞれ、100dpi，170dpi，200dpi，300dpi，400dpiのディスプレイまたは印刷物を１０インチ離れて見たときの視距離を意味する。
【００９９】
図８〜図１０に示す数値を用いて、上式（７）の重み付け係数の平方根（Ｗ_b[i] ^csf・Ｇ_b[i]）^1/2を計算した。その計算結果を、以下の表７〜表１８に示す。表７〜表９は、図８に示す数値を用いて計算した（９，７）フィルタの白黒用数値、表１０〜表１２は、図９と図１０に示す数値を用いて計算した（９，７）フィルタのカラー用数値を、表１３〜表１５は、図８に示す数値を用いて計算した（５，３）フィルタの白黒用数値を、表１６〜表１８は、図９と図１０に示す数値を用いて計算した（５，３）フィルタのカラー用数値をそれぞれ示している。
【０１００】
【表７】

【０１０１】
【表８】

【０１０２】
【表９】

【０１０３】
【表１０】

【０１０４】
【表１１】

【０１０５】
【表１２】

【０１０６】
【表１３】

【０１０７】
【表１４】

【０１０８】
【表１５】

【０１０９】
【表１６】

【０１１０】
【表１７】

【０１１１】
【表１８】

【０１１２】
次に、上記表７〜表１８に示す数値を用いて、上記第２実施例で述べたのと同じ手順で各帯域成分の優先度を算出した。すなわち、最も高い分解レベルｎの最低域成分ＬＬｎの数値をαとし、その他の帯域成分の数値をｘとし、２の巾乗に丸め込む変数ｙに関する関数をR［ｙ］とし、変数ｙの２の巾乗２^mの指数ｍを算出する関数をｍ＝Ｉ［２^m］とし、変数ｙに関する絶対値を｜ｙ｜とするとき、優先度ｐは、ｐ＝｜Ｉ［Ｒ［ｘ／α］］｜、に従って算出される。
【０１１３】
優先度の値を、以下の表１９〜表３０に示す。表１９，表２０，表２１，表２２，表２３，表２４，表２５，表２６，表２７，表２８，表２９および表３０の優先度は、それぞれ、上記した表７，表８，表９，表１０，表１１，表１２，表１３，表１４，表１５，表１６，表１７および表１８の数値を用いて算出されたものである。
【０１１４】
【表１９】

【０１１５】
【表２０】

【０１１６】
【表２１】

【０１１７】
【表２２】

【０１１８】
【表２３】

【０１１９】
【表２４】

【０１２０】
【表２５】

【０１２１】
【表２６】

【０１２２】
【表２７】

【０１２３】
【表２８】

【０１２４】
【表２９】

【０１２５】
【表３０】

【０１２６】
本実施例では、上記第２実施例と同じように、以上の表１９〜表３０に示す優先度のビット数だけ右シフトさせることで、各帯域成分の変換係数に対して優先度が設定される。これにより、人間の視覚特性を考慮した優先度を設定できる。
【０１２７】
画質制御処理．
次に、図１に示した画質制御部２３の構成と処理内容について説明する。図１１は、この画質制御部２３の概略構成を示す機能ブロック図である。
【０１２８】
この画質制御部２３は、外部から供給される目標画質情報（高画質，標準画質，低画質，解像度情報など）に基づいて、複数の画質パラメータ群から当該目標画質情報に適した画質パラメータＱＰを選択して出力する画質パラメータ選択部３１と、符号化対象を決定する符号化対象判定部３０とを備えている。符号化対象判定部３０は、優先度テーブル２４から取得した優先度データＰＤに従って、符号化データＢＤの各帯域成分に対して上述の優先度を設定する。また符号化対象判定部３０は、設定した優先度に従って、前記画質パラメータＱＰで指定される目標画質に合わせて符号化対象を決定し、画質制御信号ＣＳ１を生成出力する。
【０１２９】
以下、符号化対象の決定方法について説明する。図１２は、優先度に応じてビットシフトされた変換係数３３，３３，…を例示する模式図である。各変換係数３３は優先度に応じてビットシフトされている。また、変換係数３３の各ビットに付した番号０，１，…，１０は、当該ビットが属するビットプレーンの番号を示している。ここで、ＬＳＢ番号＝０，ＭＳＢ番号＝１０、である。
【０１３０】
符号化対象判定部３０は、画質パラメータＱＰに従って符号化終了ライン３２を設定し、当該符号化終了ライン３２よりも上位ビットを符号化対象に決定し、そのライン３２よりも下位ビットを符号化対象から外すように画質制御信号ＣＳ１を生成する。これにより符号化対象を効率的に選別することが可能になる。この結果、画質制御信号ＣＳ１を受けた算術符号化部２１は、各コードブロックにおいて、符号化終了ライン３２よりも上位のビットプレーンのみを算術符号化し、そのライン３２よりも下位のビットプレーンを切り捨てることになる。尚、算術符号化部２１は、ビットシフト処理によりゼロが挿入されたビットに対しては、算術符号化を施さない。
【０１３１】
符号化対象判定部３０は、更に、画質パラメータＱＰに従って、符号化パス単位で符号化対象を決定することができる。画質パラメータＱＰは、符号化対象のビットプレーンの制限と、符号化対象の符号化パス（ＣＬパス，ＳＩＧパスおよびＭＲパス）の制限とを示すパラメータ群を含んでいる。以下の表３１に、２０４８×２５６０画素の解像度をもつ画像に適した画質パラメータＱＰを例示する。尚、最低域のサブバンドの解像度を１２８×１２８画素よりも小さくする必要があるため、５以上の分解レベルが必要である。
【０１３２】
【表３１】

【０１３３】
表３１において「ビットプレーン数」は、図１２に示した符号化終了ライン３２よりも下位ビットの切り捨て対象のビットプレーンの数を、「パス名」は、符号化対象の中の最終符号化パスを、「最大パス数」は、符号化対象の符号化パス数の上限をそれぞれ表している。
【０１３４】
図１２と表３１を適用した場合の処理例を以下に説明する。図１３に、帯域成分ＬＬ５の変換係数３３として"00011010111₂=215₁₀"を例示する（Y₂は２進値Yを、X₁₀は１０進値Xを表すものとする）。表３１に示す通り、帯域成分ＬＬ５における最終符号化パスはＣＬパス、最大パス数は１７に制限されている。
【０１３５】
図１３に示す変換係数の７番目ビットは、ＳＩＧパスまたはＣＬパスに属するようにコンテクスト判定がなされている。８番目〜１０番目の上位ビットは、０ビットのみで構成されるビットプレーンに属する場合はタグツリー（Tag tree）と称する方式で符号化され、既に符号化パスが開始している場合はＳＩＧパスまたはＣＬパスで符号化される。７番目ビットが符号化開始パス（ＣＬパス）に属する場合、６番目ビットを含む下位ビットは、ＭＲパスに属するようにコンテクスト判定される。一般に、符号化開始のビットプレーンよりも下位のビットプレーンは、符号化効率の観点から、ＳＩＧパス，ＭＲパスおよびＣＬパスの順番で符号化される。よって、最大パス数は１７に制限されているため、７番目ビットのＣＬパスから１番目ビットのＳＩＧパス迄の計１７パスが符号化対象になる。但し、１番目ビットはＭＲパスに属するため符号化されない。従って、算術符号化処理において、下位２ビットは切り捨てられ、符号化後の値は"00011010100₂=212₁₀"となる。この値がミッドポイントで逆量子化されれば、"00011010110₂=214₁₀"となる。
【０１３６】
次に、図１４に、帯域成分ＬＬ５の変換係数３３として"00000001111₂=15₁₀"を例示する。変換係数の３番目ビットは、ＳＩＧパスまたはＣＬパスに属する。４番目〜１０番目の上位ビットは、０ビットのみで構成されるビットプレーンに属する場合はタグツリー（Tag tree）で符号化され、既に符号化パスが開始している場合はＳＩＧパスまたはＣＬパスで符号化される。３番目ビットが符号化開始パス（ＣＬパス）に属する場合、２番目ビットを含む下位ビットはＭＲパスに属し、３番目ビットのＣＬパスから０番目ビットのＣＬパス迄の計１０パスが符号化対象になる。算術符号化処理において符号化後の値は"00000001111₂=15₁₀"となり、この値が逆量子化されれば、"00000001111₂=15₁₀"となる。
【０１３７】
次に、図１５に、帯域成分ＨＨ２の変換係数３３として"00001011111₂=95₁₀"を例示する。表３１に示す通り、帯域成分ＨＨ２における最終符号化パスはＳＩＧパス、最大パス数は１４に制限されている。また下位３ビットのビットプレーンは切り捨てられる。
【０１３８】
変換係数の６番目ビットは、ＳＩＧパスまたはＣＬパスに属する。７番目〜１０番目の上位ビットは、０ビットのみで構成されるビットプレーンに属する場合はタグツリー（Tag tree）で符号化され、既に符号化パスが開始している場合はＳＩＧパスまたはＣＬパスで符号化される。６番目ビットが符号化開始パス（ＣＬパス）に属する場合、５番目ビットを含む下位ビットはＭＲパスに属する。また、３番目ビットプレーンのＳＩＧパス迄しか符号化しないという制限のため、６番目ビットのＣＬパスから４番目ビットのＳＩＧパス迄の８パスが符号化対象になるが、３番目ビットはＭＲパスに属するため符号化されない。従って、算術符号化処理において符号化後の値は"00001010000₂=80₁₀"となり、この値がミッドポイントで逆量子化されれば、"00001011000₂=88₁₀"となる。
【０１３９】
尚、各ビットプレーンを、ＳＩＧパス，ＭＲパスおよびＣＬパスの順番で符号化するのは、ＳＩＧパスの歪みに対する符号化効率が最も高いからである。図１６に、各符号化パスにおけるレート・歪み特性を示す。Ｒ−Ｄ曲線中、点Ｐ₁〜Ｐ₂の部分がＳＩＧパス，点Ｐ₂〜Ｐ₃の部分がＭＲパス、点Ｐ₃〜Ｐ₄の部分がＣＬパスを示している。各符号化パスにおけるレート（符号量）に対する歪みの比率ΔＤ_SIG／ΔＲ_SIG，ΔＤ_MR／ΔＲ_MR，ΔＤ_CL／ΔＲ_CLをみれば、ＳＩＧパスにおける曲線勾配が最も急であり、符号化効率が最も高いことが分かる。
【０１４０】
以上のように、本実施形態に係る画質制御方法では、優先度に応じてビットシフトした変換係数に対して、変換係数を符号化対象とするか否かが決定される。符号化対象のみが算術符号化部２１で選択的に算術符号化されるため、歪みの少ない高画質の圧縮画像を生成し得るように、符号量を効率良く制御することが可能である。
【０１４１】
符号量制御処理．
次に、図１に示した符号量制御部２２の構成と処理内容について説明する。図１７は、この符号量制御部２２の概略構成を示す機能ブロック図である。
【０１４２】
この符号量制御部２２は、大容量記憶装置４０，符号量算出部４１およびデータ出力制御部４２を備えている。
【０１４３】
上述したように図１に示した算術符号化部２１は、画質制御部２３から指定された符号化対象のみを選択的に算術符号化し、その結果得た符号化データＡＤを符号量制御部２２に出力する。符号量算出部４１は、入力する符号化データＡＤの容量の小計を、帯域成分単位，ビットプレーン単位および符号化パス単位で算出し、その結果得られる小計情報４３をデータ出力制御部４２に出力する。また、符号化データＡＤは、フレーム単位或いはサブフレーム単位で大容量記憶装置４０に一時記憶させられる。
【０１４４】
前記データ出力制御部４２は、大容量記憶装置４０に一時記憶された符号化データＡＤを読み出し、優先度データＰＤ２を用いて上記第１実施例〜第３実施例のようにビットシフトさせる。次いで、データ出力制御部４２は、ビットシフトした符号化データを以下に説明する走査順序で並べ替えて生成した符号列から、目標符号量に適合するように切り捨て点（truncation point）を算出する。次に、データ出力制御部４２は、その符号列のうち切り捨て点よりも前の符号列を符号化データＣＤとしてビットストリーム生成部１７に出力する。
【０１４５】
図１８と図１９は、前記走査順序と切り捨て点の一例を説明するための図である。図１８と図１９には、図１２に示したのと同じ規則で、優先度に応じてビットシフトされた変換係数３３，３３，…が表示されている。符号化終了ライン３２よりも上位ビット（図面左側のビット）の符号化データＡＤが、符号量制御部２２に入力する。
【０１４６】
図１８の矢印に示すように、変換係数３３，３３，…は、ビットプレーン単位または符号化パス単位で、優先度の高い順に（上位ビットから下位ビットに向けて）且つ同一の優先度においては高域側から低域側に向けた走査順序で並べ替えられる。一般に、下位のビットプレーンを符号化する程にＭＲパスの割合が増えて圧縮効率が下がる傾向にある。よって、圧縮効率を向上させるために出来るだけ多くのＳＩＧパスを符号化すべく、同一の優先度においては高域側から低域側に向けた走査順序を採用している。
【０１４７】
そして、データ出力制御部４２は、次式（８）に示すように、実際の符号量（バイト数）が目標符号量（バイト数）以下になる条件を満たすように切り捨て点を決定し、当該切り捨て点以降の符号列に含まれる下位ビットプレーンを切り捨てる。これにより、既に算術符号化済みの符号化データのレート制御を、各サブバンドに設定した優先度に従って効率的に行うことができる。
【０１４８】
【数８】

【０１４９】
図１９に示すように、目標符号量に合わせてサブバンドＨＬ３の２番目ビットプレーンが切り捨て点として決定された場合、矢印で示す部分のビットが切り捨てられることになる。
【０１５０】
図２０は、ビットプレーン単位で並べ替えられた符号列を示す図、図２１は、符号化パス単位で並べ替えられた符号列を示す図である。図２０では、各ビットプレーンに対して、サブバンドを示す符号ＬＬ５，ＨＬ５，…と、ビットプレーン番号１０，９，…とが付されている。サブバンドＨＬ３の２番目ビットプレーンに付されたライン４４以降のビットプレーンが切り捨てられる。
【０１５１】
また、図２１では、各符号化パスに対して、符号化パスの種類を示す符号ＣＬ，ＳＩＧ，ＭＲと、サブバンドを示す符号ＬＬ５，ＨＬ５，…と、ビットプレーン番号１０，９，…とが付されている。サブバンドＨＬ３の２番目ビットプレーンのＭＲパスに付されたライン４４以降のビットプレーンが切り捨てられる。
【０１５２】
以上のように本実施形態に係る符号量制御処理によれば、レート・歪み最適化処理のために各符号化パスにおける歪量を算出せずに済むため、リアルタイム性が高く、オーバーヘッドが大幅に低減した高効率のレート制御を実現できる。
【０１５３】
【発明の効果】
以上の如く、本発明の請求項１に係る圧縮符号化装置、請求項９に係る圧縮符号化方法および請求項１７に係るプログラムによれば、上記帯域成分は、それぞれ、低域成分に再帰的に帯域分割された回数に応じて優先度を設定され、この優先度に応じて符号化対象になるか否かが決定される。符号化対象のみが選択的にエントロピー符号化されるため、符号量を効率的に制御でき、少ない演算量で高速な符号化処理を行うことが可能になる。
【０１５４】
請求項２、請求項１０および請求項１８によれば、人間の視覚評価に適した、高い表示画質を有する圧縮画像を生成することが可能となる。
【０１５５】
請求項３、請求項１１および請求項１９によれば、各帯域成分の変換係数は優先度に対応するビット数だけシフトされ、シフトするビット数に応じて変換係数の優先度が定まることから、圧縮画像の画質に応じて符号化対象を効率的に指定することが可能になる。
【０１５６】
請求項４，５，６、請求項１２，１３，１４および請求項２０，２１，２２によれば、ビットプレーン単位および符号化パス単位で、優先度に応じて符号化対象を細かく指定できることから、符号量を正確に制御し、圧縮画像の画質を細かく制御することが可能となる。
【０１５７】
請求項７，８、請求項１５，１６および請求項２３，２４によれば、既にエントロピー符号化されたデータのレート制御を、各帯域成分に設定した優先度に従って効率的に行うことが可能である。また、従来のようにレート・歪み最適化を用いた最適化処理を行わなくても、歪みを抑制し得るようにリアルタイム性の高いレート制御を行うことが可能である。
【図面の簡単な説明】
【図１】本発明の実施形態に係る圧縮符号化装置の概略構成を示す機能ブロック図である。
【図２】ウェーブレット変換によって帯域分割した２次元画像を示す模式図である。
【図３】ビットシフトによる優先度設定処理を説明するための図である。
【図４】ビットシフトされた変換係数を例示する図である。
【図５】ウェーブレット変換によって帯域分割した２次元画像を示す模式図である。
【図６】ウェーブレット変換によって帯域分割した２次元画像を示す模式図である。
【図７】図６に示す優先度に応じて右ビットシフトされた帯域成分の変換係数を示す模式図である。
【図８】 Energy weighting factorの数値テーブルを示す図である。
【図９】 Energy weighting factorの数値テーブルを示す図である。
【図１０】 Energy weighting factorの数値テーブルを示す図である。
【図１１】本実施形態に係る画質制御部の概略構成を示す機能ブロック図である。
【図１２】優先度に応じてビットシフトされた変換係数を例示する模式図である。
【図１３】帯域成分ＬＬ５の変換係数の符号化処理例を説明するための図である。
【図１４】帯域成分ＬＬ５の変換係数の符号化処理例を説明するための図である。
【図１５】帯域成分ＨＨ２の変換係数の符号化処理例を説明するための図である。
【図１６】レート・歪み特性の曲線を示す図である。
【図１７】本実施形態に係る符号量制御部の概略構成を示す機能ブロック図である。
【図１８】走査順序の一例を説明するための図である。
【図１９】切り捨て点の一例を説明するための図である。
【図２０】ビットプレーン単位で並べ替えられた符号列を示す図である。
【図２１】符号化パス単位で並べ替えられた符号列を示す図である。
【図２２】ＪＰＥＧ２０００方式による圧縮符号化装置の概略構成を示す機能ブロック図である。
【図２３】オクターブ分割方式に従って帯域分割された２次元画像を示す模式図である。
【図２４】複数のコードブロックに分解された２次元画像を示す模式図である。
【図２５】コードブロックを構成する複数枚のビットプレーンを示す模式図である。
【図２６】３種類の符号化パスを示す模式図である。
【図２７】レートと歪みの関係を表すＲ−Ｄ曲線を示す図である。
【符号の説明】
１ 圧縮符号化装置
１０ ＤＣレベルシフト部
１１ 色空間変換部
１２ タイリング部
１３ ＤＷＴ部
１４ 量子化部
１５ ＲＯＩ部
１７ ビットストリーム生成部
２０ 係数ビットモデリング部
２１ 算術符号化部
２２ 符号量制御部
２３ 画質制御部
２４ 優先度テーブル[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a compression encoding method and a compression encoding apparatus used in an image compression / decompression technique.
[0002]
[Prior art]
As a next-generation high-efficiency encoding system for image data, the ISO (International Organization for Standardization) and ITU-T (International Telecommunication Union Telecommunication Standardization Sector) have developed the JPEG 2000 (Joint Photographic Experts Group 2000) system. The JPEG2000 system has superior functions compared with the current mainstream JPEG (Joint Photographic Experts Group) system, adopts DWT (Discrete Wavelet Transform) as orthogonal transform, and uses bit for entropy coding. It is characterized in that a method called EBCOT (Embedded Block Coding with Optimized Truncation) that performs plain coding is adopted.
[0003]
FIG. 22 is a functional block diagram showing a schematic configuration of an image compression coding apparatus based on the JPEG2000 system. Hereinafter, the compression encoding procedure of the JPEG2000 system will be outlined with reference to FIG.
[0004]
The image signal input to the compression encoding apparatus 100 is subjected to DC level conversion as necessary by the DC level shift unit 102 and then output to the color space conversion unit 103. Next, the color space conversion unit 103 converts the color space of the signal input from the DC level shift unit 102. Here, for example, the RGB signal input to the color space conversion unit 103 is converted into a YCbCr signal (a signal composed of a luminance signal Y and color difference signals Cb and Cr).
[0005]
Next, the tiling unit 104 divides the image signal input from the color space conversion unit 103 into a plurality of area components called “tiles” having a rectangular shape, and outputs the result to the DWT unit 105. The DWT unit 105 performs integer type or real number type DWT on the image signal input from the tiling unit 104 in units of tiles, and outputs a conversion coefficient obtained as a result. In the DWT, a one-dimensional filter that divides a two-dimensional image signal into a high frequency component (high frequency component) and a low frequency component (low frequency component) is applied in the order of vertical and horizontal directions. The JPEG 2000 basic method employs an octave division method that recursively divides a band only in a band component that is divided in the vertical direction and the horizontal direction in the low frequency range. The number of times the band is recursively divided is called a decomposition level.
[0006]
FIG. 23 is a schematic diagram showing a two-dimensional image 120 that has been subjected to decomposition level 3 DWT according to the octave division method. At decomposition level 1, the two-dimensional image 120 is divided into four band components HH1, HL1, LH1, and LL1 (not shown) by sequentially applying the above-described one-dimensional filter in the vertical direction and the horizontal direction. The Here, “H” indicates a high frequency component, and “L” indicates a low frequency component. For example, HL1 is a band component composed of a high frequency component H in the horizontal direction and a low frequency component L in the vertical direction at the decomposition level 1. Generalizing the notation, “XYn” (X and Y are either H or L; n is an integer equal to or greater than 1) is defined as the horizontal band component X and the vertical band component Y at the decomposition level n. A band component consisting of
[0007]
At decomposition level 2, the low frequency component LL1 is band-divided into HH2, HL2, LH2, and LL2 (not shown). Further, at the decomposition level 3, the low frequency component LL2 is band-divided into HH3, HL3, LH3 and LL3. FIG. 23 shows the arrangement of the band components HH1 to LL3 generated as described above. In FIG. 23, an example of the third-order decomposition level is shown, but in the JPEG2000 system, generally, third-order to eighth-order decomposition levels are employed.
[0008]
Next, the quantization unit 106 has a function of performing scalar quantization on the transform coefficient output from the DWT unit 105 as necessary. The quantization unit 106 also has a function of performing bit shift processing that prioritizes the image quality of a designated region (ROI; Region Of Interest) by the ROI unit 107. Note that, when performing reversible (lossless) transformation, scalar quantization in the quantization unit 106 is not performed. In the JPEG2000 system, two types of quantization means are prepared: scalar quantization in the quantization unit 106 and post-quantization (truncation) described later.
[0009]
Next, the transform coefficient output from the quantization unit 106 is sequentially subjected to block-based entropy coding by the coefficient bit modeling unit 108 and the arithmetic coding unit 109 in accordance with the above-described EBCOT, and the code amount control unit 110. The rate is controlled by. Specifically, the coefficient bit modeling unit 108 divides the band component of the input transform coefficient into areas called “code blocks” of about 16 × 16, 32 × 32, and 64 × 64, and further each code block. , It is decomposed into a plurality of bit planes composed of a two-dimensional array of bits.
[0010]
FIG. 24 is a schematic diagram showing a two-dimensional image 120 decomposed into a plurality of code blocks 121, 121, 121,. FIG. 25 shows n bit planes 122 constituting the code block 121.₀~ 122_n-1It is a schematic diagram which shows (n: natural number). As shown in FIG. 25, when the binary value 123 of the conversion coefficient at one point in the code block 121 is “011... 0”, the bits constituting the binary value 123 are respectively bit planes 122._n-1122_n-2122_n-3, ..., 122₀It is disassembled to belong to. Bit plane 122 in the figure_n-1Represents the most significant bit plane consisting only of the most significant bit (MSB) of the transform coefficient, and represents the bit plane 122.₀Represents the least significant bit plane consisting only of the least significant bit (LSB).
[0011]
  Further, the coefficient bit modeling unit 108 includes each bit plane 122._kThe context of each bit in (k = 0 to n−1) is determined, and as shown in FIG. 26, the bit plane 122 is determined according to the significance (determination result) of each bit._kWith three types of encoding passes:SIG pass ( SIGnificance propagation pass ), MR path ( Magnitude Refinement pass ), CL path ( CLeanup pass )Disassembled into The context determination algorithm for each coding pass is defined by EBCOT. According to it, “significant” means a state in which the coefficient of interest is known to be non-zero in the encoding process so far, and “not significant” means that the coefficient value is zero. , Or a state that may be zero.
[0012]
The coefficient bit modeling unit 108 performs an SIG path (significant coefficient coding pass with significant coefficients around), an MR path (significant coefficient coding path), and a CL path (SIG path, MR path not corresponding to the rest). Bit plane encoding is performed in three types of encoding passes). Bit plane encoding is performed by scanning the bits of each bit plane in units of 4 bits from the most significant bit plane to the least significant bit plane and determining whether or not a significant coefficient exists. The number of bit planes composed only of insignificant coefficients (0 bits) is recorded in the packet header, and actual encoding is started from the bit plane in which significant coefficients first appear. The encoding start bit plane is encoded only by the CL pass, and the bit planes lower than the bit plane are sequentially encoded by the above three types of encoding passes.
[0013]
FIG. 27 shows an RD curve representing the relationship between rate (code amount; R) and distortion (D). In this RD curve, R₁Is the rate before bit-plane coding, R₂Is the bit-plane encoded rate, D₁Is the distortion before bit-plane coding, D₂Indicates distortion after bit-plane encoding. A, B, and C are labels representing the above-described encoding pass. For efficient coding, the starting point P₁(R₁, D₁) To end point (R₂, D₂), It is preferable to adopt a concave-curved ABC path rather than a convex-curved CBA path. In order to realize such a concave curve, it is known that encoding may be performed from the MSB plane toward the LSB plane.
[0014]
Next, the arithmetic coding unit 109 performs arithmetic coding in units of coding passes on the coefficient sequence from the coefficient bit modeling unit 108 based on the context determination result using the MQ coder. There is also a mode in which the arithmetic encoding unit 109 performs a bypass process in which a part of the coefficient sequence input from the coefficient bit modeling unit 108 is not arithmetically encoded.
[0015]
Next, the code amount control unit 110 controls the final code amount by performing post-quantization that truncates the lower bit planes of the code string output by the arithmetic encoding unit 109. Then, the bit stream generation unit 111 generates a bit stream in which the code string output from the code amount control unit 110 and additional information (header information, layer configuration, scalability information, quantization table, etc.) are multiplexed, and a compressed image Output as.
[0016]
[Problems to be solved by the invention]
Conventionally, a method called rate / distortion optimization (R-D optimization) has been adopted as a rate control method in the code amount control unit 110. The algorithm for rate / distortion optimization is disclosed in the document "David S. Taubman and Michael W. Marcellin," JPEG2000 IMAGE COMPRESSION FUNDAMENTALS, STANDARDS AND PRACTICE, "Kluwer Academic Publishers" (hereinafter referred to as Reference A). Has been.
[0017]
However, in this method, (1) it is necessary to calculate the distortion amount with respect to the rate one by one in each coding pass, and it is necessary to estimate the optimal solution at a certain coding rate, which increases the amount of calculation and real time. There is a problem that (2) a memory for storing the distortion amount calculated in each encoding pass is required.
[0018]
In view of the above problems and the like, an object of the present invention is to provide a compression encoding apparatus and a compression encoding method capable of executing an encoding process capable of suppressing distortion with respect to an encoding rate with a small amount of computation and at a high speed. The point is to provide.
[0019]
[Means for Solving the Problems]
  In order to solve the above-mentioned problem, the invention according to claim 1 is a compression encoding apparatus for compressing and encoding an image signal, and recursively divides the image signal into a high frequency component and a low frequency component by wavelet transform. A wavelet transform unit that generates and outputs transform coefficients of a plurality of band components; an entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;Based on the norm of the two-dimensional synthesis filter coefficient of the predetermined band component among the plurality of band components, andDepending on the number of times the band was recursively splitOf each band componentAn image quality control unit that sets priority and determines the encoding target to be specified in the entropy encoding unit according to the priority is provided.
[0020]
  The invention according to claim 2Compress and encode image signalsA compression encoding device comprising:A wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components, and a specified encoding target among the transform coefficients An entropy encoding unit that selectively entropy-encodes only the band, a priority is set according to the number of times each band component is recursively band-divided into the low-frequency components, and the entropy encoding unit An image quality control unit that determines the encoding target to be specified in accordance with the priority,The image quality control unit has a function of using the priority weighted in consideration of human visual characteristics.
[0021]
The invention according to claim 3 is the compression encoding apparatus according to claim 1 or 2, wherein the image quality control unit is obtained by shifting the transform coefficient of each band component by the number of bits corresponding to the priority. The encoding target is determined from the transformed coefficients.
[0022]
The invention according to claim 4 is the compression encoding apparatus according to any one of claims 1 to 3, wherein the transform coefficient output from the wavelet transform unit is set to 2 for each bit of the transform coefficient. A coefficient bit modeling unit that decomposes the bit plane into a plurality of bit planes configured in a dimensional array and generates encoded data by decomposing the bit plane into an encoding pass determined according to the significance of each bit. Further, the entropy encoding unit selectively entropy encodes only the designated encoding target among the encoded data generated by the coefficient bit modeling unit.
[0023]
The invention according to claim 5 is the compression encoding apparatus according to claim 4, wherein the image quality control unit has a function of determining the encoding target in units of the bit planes.
[0024]
The invention according to claim 6 is the compression encoding apparatus according to claim 4, wherein the image quality control unit has a function of determining the encoding target in units of the encoding pass.
[0025]
The invention according to claim 7 is the compression encoding apparatus according to any one of claims 4 to 6, wherein the code amount control unit controls the rate of the encoded data output from the entropy encoding unit. The code amount control unit calculates a subtotal of the capacity of the encoded data output from the entropy encoding unit in at least one unit of the band component, the bit plane, and the encoding pass. And a subtotal calculated by the code amount calculation unit, and a truncation point that matches the target code amount is calculated from a code string generated by rearranging the encoded data in a predetermined scanning order. And a data output control unit that outputs the code string before the cut-off point.
[0026]
The invention according to claim 8 is the compression encoding apparatus according to claim 7, wherein the data output control unit sets the transform coefficients in order from the highest priority to the higher frequency side in the same priority. The code sequence is generated by rearranging in the scanning order from the low frequency side to the low frequency side.
[0027]
  Next, an invention according to claim 9 is a compression encoding method for compressing and encoding an image signal, and (a) the image signal is recursively divided into a high frequency component and a low frequency component by wavelet transform. A step of generating transform coefficients of a plurality of band components; (b) a step of selectively entropy-encoding only a specified encoding target among the transform coefficients; and (c).Based on the norm of the two-dimensional synthesis filter coefficient of the predetermined band component among the plurality of band components, andDepending on the number of times the band was recursively split,And setting a priority, and determining the encoding target specified in the step (b) according to the priority.
[0028]
  The invention according to claim 10 is:Compress and encode image signalsA compression encoding method comprising:(A) a step of recursively dividing the image signal into a high-frequency component and a low-frequency component by wavelet transform to generate conversion coefficients of a plurality of band components; and (b) a specified one of the conversion coefficients. A step of selectively entropy encoding only the encoding target; and (c) setting a priority according to the number of times each band component is recursively band-divided into the low-frequency components, and the step Determining the encoding target specified in (b) according to the priority,The step (c) includes the step of determining the priority weighted in consideration of human visual characteristics.
[0029]
The invention according to claim 11 is the compression encoding method according to claim 9 or 10, wherein the step (c) is performed by shifting the transform coefficient of each band component by the number of bits corresponding to the priority. This is a step of determining the encoding target from the obtained transform coefficient.
[0030]
The invention according to claim 12 is the compression encoding method according to any one of claims 9 to 11, wherein after the step (a) and before the step (b), (d) the The transform coefficient generated in the step (a) is decomposed into a plurality of bit planes configured by two-dimensionally arranging each bit of the transform coefficient, and the bit plane is determined according to the significance of each bit. Further comprising a step of generating encoded data by decomposing into the determined encoding pass, and the step (b) is the specified code among the encoded data generated in the step (d). This is a process of selectively entropy encoding only the conversion target.
[0031]
A thirteenth aspect of the present invention is the compression encoding method according to the twelfth aspect, wherein the step (c) includes a step of determining the encoding target in units of the bit planes.
[0032]
The invention according to claim 14 is the compression encoding method according to claim 12, wherein the step (c) includes a step of determining the encoding target in units of the encoding pass.
[0033]
The invention according to claim 15 is the compression encoding method according to any one of claims 12 to 14, wherein (e) the rate of the encoded data subjected to entropy encoding in the step (b) is controlled. The step (e) is entropy-encoded in the step (b) in (e-1) at least one unit of the band component, the bit plane, and the coding pass. A step of calculating a subtotal of the capacity of the encoded data; and (e-2) a code generated by rearranging the encoded data in a predetermined scanning order using the subtotal calculated in the step (e-1). And calculating a truncation point that matches the target code amount from the sequence and outputting the code sequence before the truncation point.
[0034]
The invention according to claim 16 is the compression encoding method according to claim 15, wherein the step (e-2) sets the transform coefficients in the descending order of priority and in the same priority. Rearrangement in the scanning order from the band side toward the low band side to generate the code string.
[0035]
  Next, an invention according to claim 17 is a program for causing a microprocessor to compress and encode an image signal, and recursively divides the image signal into a high frequency component and a low frequency component by wavelet transform. A wavelet transform unit that generates and outputs transform coefficients of a plurality of band components; an entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;Based on the norm of the two-dimensional synthesis filter coefficient of the predetermined band component among the plurality of band components, andDepending on the number of times the band was recursively split,The microprocessor is made to function as an image quality control unit that sets priority and determines the encoding target to be specified in the entropy encoding unit according to the priority.
[0036]
  The invention according to claim 18To have a microprocessor compress and encode an image signalThe program ofA wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components, and a specified encoding target among the transform coefficients An entropy encoding unit that selectively entropy-encodes only the band, and a priority is set according to the number of times the band components are recursively band-divided into the low-frequency components, and the entropy encoding unit The microprocessor functions as an image quality control unit that determines the encoding target to be specified according to the priority,The image quality control unit causes the microprocessor to function so as to use the priority weighted in consideration of human visual characteristics.
[0037]
The invention according to claim 19 is the program according to claim 17 or 18, wherein the image quality control unit shifts the conversion coefficient of each band component by the number of bits corresponding to the priority. The microprocessor is caused to function so as to determine the encoding target.
[0038]
An invention according to claim 20 is the program according to any one of claims 17 to 19, wherein the transform coefficient output from the wavelet transform unit is arranged in a two-dimensional array with each bit of the transform coefficient. The microprocessor as a coefficient bit modeling unit that generates encoded data by decomposing the bit plane into a plurality of encoding planes determined according to the significance of each bit. And the entropy coding unit causes the microprocessor to selectively entropy-encode only the designated coding target among the coded data generated by the coefficient bit modeling unit. It is to function.
[0039]
The invention according to claim 21 is the program according to claim 20, wherein the image quality control section causes the microprocessor to function so as to determine the encoding target in units of the bit planes.
[0040]
The invention according to claim 22 is the program according to claim 20, wherein the image quality control unit causes the microprocessor to function so as to determine the encoding target in units of the encoding pass.
[0041]
The invention according to claim 23 is the program according to any one of claims 20 to 22, wherein the micro is used as a code amount control unit for controlling a rate of encoded data output from the entropy encoding unit. The code amount control unit causes the processor to function and the subtotal of the capacity of the encoded data output from the entropy encoding unit in at least one unit of the band component, the bit plane, and the encoding path. Using the code amount calculation unit for calculating the code amount and the subtotal calculated by the code amount calculation unit, a truncation point that matches the target code amount is generated from a code string generated by rearranging the encoded data in a predetermined scanning order. The microprocessor is caused to function as a data output control unit that calculates and outputs the code string before the cut-off point.
[0042]
The invention according to claim 24 is the program according to claim 23, wherein the data output control unit sets the conversion coefficients in order from the highest priority to the lower frequency in the same priority. The microprocessor is caused to function so as to generate the code string by rearranging in the scanning order toward the side.
[0043]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described.
[0044]
Compression encoding device.
FIG. 1 is a functional block diagram showing a schematic configuration of a compression coding apparatus 1 according to an embodiment of the present invention. After an overview of the configuration and function of the compression coding apparatus 1, the coding block and the coding method according to the present embodiment will be described in detail.
[0045]
The compression coding apparatus 1 includes a DC level shift unit 10, a color space conversion unit 11, a tiling unit 12, a DWT unit 13, a quantization unit 14, an ROI unit 15, a coefficient bit modeling unit 20, an arithmetic coding unit (entropy). (Encoding unit) 21, code amount control unit 22, image quality control unit 23, priority table 24, and bit stream generation unit 17.
[0046]
Note that all or part of each of the processing units 10 to 15, 17, and 20 to 24 constituting the compression encoding device 1 may be configured by hardware or a program that causes a microprocessor to function. May be.
[0047]
The image signal input to the compression encoding device 1 is subjected to DC level conversion as necessary by the DC level shift unit 10 and then output to the color space conversion unit 11. The color space conversion unit 11 performs color space conversion on the input signal and outputs it. According to the JPEG2000 system, RCT (Reversible Component Transformation) for reversible conversion and ICT (Irreversible Component Transformation) for irreversible conversion are prepared as color space conversion, and one of them can be selected as appropriate. Thereby, for example, input RGB signals are converted into YCbCr signals or YUV signals.
[0048]
Next, the tiling unit 12 divides the image signal input from the color space conversion unit 11 into a plurality of region components called “tiles” having a rectangular shape, and outputs the region components to the DWT unit 13. It is not always necessary to divide the image signal into tiles, and the image signal for one frame may be output to the next functional block as it is.
[0049]
Next, the DWT unit 13 performs integer type or real type DWT on the image signal input from the tiling unit 12 in units of tiles, thereby converting the image signal into a high frequency component and a low frequency component according to the octave division method. And band recursively. As a result, transform coefficients of a plurality of band components (subbands) HH 1 to LL 3 as shown in FIG. 23 are generated and output to the quantization unit 14. Specifically, for real type DWT, a filter of 9 × 7 tap, 5 × 3 tap or 7 × 5 type is used, and for integer type DWT, 5 × 3 tap or 13 × 7 tap, etc. The filter is used. Further, the processing of these filters may be executed by a convolution operation, or may be executed by a lifting scheme that is more efficient than the convolution operation.
[0050]
The quantization unit 14 has a function of performing scalar quantization on the transform coefficient input from the DWT unit 13. The quantization unit 14 also has a function of performing bit shift processing that prioritizes the image quality of a designated region (ROI; Region Of Interest) by the ROI unit 15. The quantization unit 14 may perform scalar quantization or may not perform scalar quantization.
[0051]
Next, the transform coefficient QD output from the quantization unit 14 is sequentially subjected to block-based entropy coding by the coefficient bit modeling unit 20 and the arithmetic coding unit 21, and the rate control unit 22 controls the rate. Is done.
[0052]
  Similar to the coefficient bit modeling unit 108 shown in FIG. 22, the coefficient bit modeling unit 20 divides the band component of the input transform coefficient QD into code blocks of about 32 × 32 or 64 × 64, and each code block Are broken down into a plurality of bit planes configured by two-dimensionally arranging each bit. As a result, each code block has a plurality of bit planes 122 as shown in FIG.₀~ 122_n-1Is broken down into The coefficient bit modeling unit 20 further performs context determination for each bit, and converts each bit plane into three types.SIG pass ( SIGnificance propagation pass ), MR pass ( Magnitude Refinement pass ) And CL path ( CLeanup pass )Are encoded by decomposing the data into the encoding passes, and the resulting encoded data BD is output.
[0053]
Next, the arithmetic encoding unit 21 arithmetically encodes only the encoding target designated by the image quality control unit 23 among the encoded data BD input from the coefficient bit modeling unit 20, and the encoded data AD obtained as a result thereof. Is output to the code amount control unit 22. Here, the arithmetic encoding unit 21 may perform bypass processing in which a part of the encoding target is not arithmetically encoded and the encoding target is included in the encoded data AD and output as it is. In addition, although this embodiment employ | adopts arithmetic encoding, you may employ | adopt not only this but entropy encoding of another system.
[0054]
Here, the image quality control unit 23 sets the priority indicating the coding order for each band component in accordance with the priority data PD acquired from the priority table 24, and the arithmetic coding unit 21 according to the priority. The encoding target specified in is determined. The priority setting method and the encoding target determination method will be described later.
[0055]
Next, the code amount control unit 22 has a function of controlling the rate of the encoded data AD input from the arithmetic encoding unit 21 using the priority data PD2 acquired from the priority table 24. That is, the code amount control unit 22 has a low priority for the encoded data AD in band component units, bit plane units, or encoding path units in accordance with the target code amount (the code amount of the final compressed image). It has a function of performing post-quantization of truncating in order. The post-quantization processing method will be described later.
[0056]
Then, the bit stream generation unit 17 generates a bit stream in which the encoded data CD output from the code amount control unit 22 and additional information (header information, layer configuration, scalability, quantization table, etc.) are multiplexed, Output to the outside as a compressed image.
[0057]
Priority setting method (first embodiment).
Next, a description will be given of a first embodiment of a method for setting the priority recorded in the priority table 24. In the present invention, the priority is determined for each band component (subband) according to the number of times the band is recursively divided into low frequency components. In this embodiment, the priority of the band component HHn at the decomposition level n (n: an integer of 1 or more) is “n−1”, the priority of the band components HLn and LHn is “(n−1) +1”, and the band component. The priority of LLn is determined as “(n−1) +2”. For example, the priority of the band component HH1 shown in FIG. 23 is set to “0”, and the priority of the band component LL3 is set to “4”. FIG. 2 is a schematic diagram showing a two-dimensional image 25 obtained by band division according to the octave division method. Each band component has a priority “0”, “1”, “2”, “3”, or “4”.
[0058]
In the priority table 24, priority information corresponding to each of the band components HHn, HLn, LHn, and LLn is recorded, and the image quality control unit 23 and the code amount control unit 22 are acquired from the priority table 24. In accordance with the priority data PD and PD2, the priority is set for each band component. Specifically, the priority is set for each transform coefficient by shifting the transform coefficient of each band component by the number of bits corresponding to the priority. In this bit shift process, it is not always necessary to actually perform a bit shift operation on each transform coefficient, and the position of each bit of the transform coefficient may be virtually shifted. In this case, the position of the bit plane to which each bit of the transform coefficient belongs does not change.
[0059]
FIG. 3 is a diagram for explaining priority setting processing by bit shift. In the example shown in FIG. 2, since the priority of the band component LL3 is “4”, the corresponding transform coefficient 26 is shifted left by 4 bits. Further, the conversion coefficients 26 and 26 of the band components HL3 and LH3 set with the priority “3” are shifted left by 3 bits, and the conversion coefficients 26 of the band components HH3, HL2 and LH2 set with the priority “2” are set. , 26, and 26 are shifted left by 2 bits, and the transform coefficients 26, 26, and 26 of the band components HH2, HL1, and LH1 set with the priority "1" are shifted left by 1 bit. At this time, as shown in FIG. 4, the conversion coefficient of the two-dimensional image 25A before the bit shift is changed to the conversion coefficient indicated by the two-dimensional image 25B by the left bit shift process. For example, the transform coefficient value (= 4) of the band component LL3 is 4 × 2 by 4-bit left shift.^Four= 64.
[0060]
As will be described later, the image quality control unit 23 can efficiently determine the encoding target instructed to the arithmetic encoding unit 21 from the bit-shifted transform coefficient sequence as shown in FIG.
[0061]
Next, the reason (theoretical background) for setting the priority as described above will be described below.
[0062]
In the above-described conventional rate / distortion optimization (R-D optimization) method, an optimization process using a distortion measure has been performed. According to the above reference A by David S. Taubman et al._i ^(z)Is calculated according to the following equation (1).
[0063]
[Expression 1]

[0064]
In the above formula (1), z is a bit truncation point; oy_i ^{K [i, j]}[J] is the jth sample value (coefficient value) of the code block dequantized by the K [i, j] th bitplane; y_i[J] is the j-th sample value (coefficient value) of the code block; G_{b [i]}Is the square of the norm of the synthesis filter coefficient corresponding to the subband b [i], and indicates the weighting coefficient of the distortion model depending on the subband b. For convenience of explanation, the symbol notation shown in the above formula (1) is slightly different from that in Reference A.
[0065]
For rate / distortion optimization, this distortion measure D_i ^(z)Optimization processing is performed so as to minimize the total amount in the subband b [i]. Weighting factor G for subband b_bRepresents weighting for reducing image distortion.
[0066]
Weighting factor G for subband b_bIs calculated according to the following equation (2).
[0067]
[Expression 2]

[0068]
Here, in the above equation (2), s_b[N] indicates a one-dimensional synthesis filter coefficient of subband b. The symbol || x || indicates a norm related to the vector x.
[0069]
According to the equations (4.39) and (4.40) described in Reference A, the one-dimensional synthesis filter coefficient s of the low-frequency component L1 at the decomposition level 1_{L [1]}[N] and the one-dimensional synthesis filter coefficient s of the high-frequency component H1 at the same decomposition level_{H [1]}[N] is calculated according to the following equation (3).
[0070]
[Equation 3]

[0071]
Here, in the above formula (3), g₀[N] is a low-pass filter coefficient of a forward transform filter for dividing an image signal into bands, g₁[N] indicates the high-pass filter coefficient, respectively.
[0072]
Further, the one-dimensional synthesis filter coefficient s of the low frequency component Ld at the decomposition level d (d = 1, 2,..., D)._{L [d]}[N] and the one-dimensional synthesis filter coefficient s of the high-frequency component Hd at the same decomposition level_{H [d]}[N] is calculated according to the following equation (4).
[0073]
[Expression 4]

[0074]
Then, the square of the norm of the one-dimensional synthesis filter coefficient of the low frequency component Ld at the decomposition level d is calculated according to the following equation (5).
[0075]
[Equation 5]

[0076]
The square of the norm of the one-dimensional synthesis filter coefficient of the high-frequency component can be calculated in the same manner as the above equation (5).
[0077]
Next, the two-dimensional synthesis filter coefficients of the band components LLD, HLd, LHd, and HHd at the decomposition level d (d = 1, 2,..., D; D is an integer) are expressed by the product of the one-dimensional synthesis filter coefficients. The two-dimensional weighting factor G of the band component b_bCan also be expressed as a product of one-dimensional weighting factors. Specifically, the two-dimensional synthesis filter coefficient and the two-dimensional weighting coefficient are calculated according to the following equation (6).
[0078]
[Formula 6]

[0079]
In the above formula (6), the subscript LL [D] indicates the subband LLD, and HL [d], LH [d], and HH [d] indicate the subbands HLd, LHd, and HHd, respectively.
[0080]
Weight coefficient G_bIs the norm. Tables 1 to 4 below show the two-dimensional weighting factor G_bThe calculation result about is shown. Table 1 shows the numerical value of the norm of each band component of the (9, 7) filter (9 × 7 tap filter), and Table 2 shows the numerical value of the norm corresponding to Table 1. Table 3 shows the norm square value of each band component of the (5, 3) filter (5 × 3 tap filter), and Table 4 shows the norm value corresponding to Table 3.
[0081]
[Table 1]

[0082]
[Table 2]

[0083]
[Table 3]

[0084]
[Table 4]

[0085]
Furthermore, when the norm of the low-frequency component LL1 at the decomposition level 1 is represented by α, a value as shown in FIG. 5 is set for each band component using this norm α. A two-dimensional image 27 shown in FIG. 5 is a diagram showing a two-dimensional image 120 that is band-divided according to the octave division method. The set value of the band component HHn at the decomposition level n (n is an integer of 1 or more) is “2^n-3× α ”, the set values of the band components HLn and LHn are“ 2 ”^n-2× α ”, the set value of the band component LLn is“ 2 ”^n-1Xα ”. Therefore, for example, the set value of the band component LH1 is “2”.^-1× α ”.
[0086]
If the set value is compared with the norm values shown in Tables 2 and 4, they are approximately approximate. For example, in the case of Table 2 (α = 1.96591), the “set value (corresponding band component)” of each band component shown in FIG. 5 is about 0.49 (HH1), about 0.98 (HL1, LH1), It is about 1.96 (HL2, LH2, HH3), about 3.93 (HL3, LH3), and about 7.86 (LL3), and these set values are close to the norm values shown in Table 2. I understand.
[0087]
In FIG. 5, the norm of the band component LL1 is rounded to α = 2, and the set value of each band component is shifted to the left by 1 bit, that is, all set values are 2¹It can be seen that the exponent of the power value of 2 multiplied by 1 matches the priority value shown in FIG. Therefore, setting the priority to each band component as in the first embodiment approximately means that the norm (square root of the weighting factor) of the filter used in the rate / distortion optimization is the sample value of each band component ( Equivalent to multiplying (conversion coefficient value). Accordingly, the priority of the present embodiment is set so as to reduce image distortion.
[0088]
Priority setting method (second embodiment).
Next, a second embodiment of the priority setting method will be described. In the present embodiment, the weight coefficient G of each band component_bA value obtained by dividing the norm, which is the square root of, by the norm of the lowest frequency component LL at the highest decomposition level is rounded to a power value of 2, and the absolute value of the exponent of the power value of 2 is set as the priority. Specifically, the norm of the lowest band component LLn of the highest decomposition level n is α, the norm of the other band component is x, a function related to a variable y rounded to the power of 2 is R [y], and the variable y 2 power of 2^mA function for calculating an index m of m = I [2^m] And the absolute value for the variable y is | y |, the priority p is calculated according to p = | I [R [x / α]] |.
[0089]
Table 5 below shows the priority calculated using the norm of the (9, 7) filter shown in Table 2 above. FIG. 6 is a band division diagram showing the two-dimensional image 28 in which the priorities shown in Table 5 are shown. Here, the highest decomposition level is 5, and α = 33.4923. In the table, “x” means that the priority of the band component is not calculated.
[0090]
[Table 5]

[0091]
Table 6 below shows the priority calculated using the norm of the (5, 3) filter shown in Table 4 above.
[0092]
[Table 6]

[0093]
In the first embodiment, the priority is set by shifting the conversion coefficient of each band component to the left by the number of priority bits, but in the second embodiment, the conversion coefficient of each band component is set to the priority. A process of shifting right by the number of bits is executed. However, the right bit shift process is executed so as to increase the bit length of the transform coefficient. FIG. 7 is a schematic diagram showing band component conversion coefficients 29, 29,... Shifted to the right by the number of priority bits shown in FIG.
[0094]
Priority setting method (third embodiment).
Next, a priority setting method according to the third embodiment in consideration of human visual characteristics will be described. When the priority shown in the second embodiment is applied to a high-resolution image of about several million pixels, the image quality of the decoded image is good for objective evaluation, but not necessarily good for human visual evaluation. Therefore, the priority setting method of this embodiment employs a weighted priority in consideration of human visual characteristics. This makes it possible to generate a compressed image with high display quality.
[0095]
Chapter 16 of the reference A describes a weighted mean squared error (WMSE) based on CSF (human visual system contrast sensitivity function). According to this description, in order to improve human visual evaluation, it is desirable to modify the above equation (1) into the following equation (7).
[0096]
[Expression 7]

[0097]
Here, in the above formula (7), W_{b [i]} ^csfIs called the “energy weighting factor” of subband b [i] and W_{b [i]} ^csfThe recommended value for is the ISO / IEC JTC 1 / SC 29 / WG1 (ITU-T SG8) N2406, "JPEG 2000 Part 1 FDIS (includes COR 1, COR 2, and DCOR3)," 4 December 2001 document ( Hereinafter referred to as Reference Document B). 8 to 10 show numerical values of “energy weighting factor” described in Reference Document B. FIG.
[0098]
8 to 10, “level” and “Lev” indicate the decomposition level, “Comp” indicates the luminance component Y and the color difference components Cb and Cr, respectively, and “Viewing distance” is 1000 and 1700. , 2000, 3000, 4000 examples are shown. "Viewing distance 1000", "Viewing distance 1700", "Viewing distance 2000", "Viewing distance 3000", and "Viewing distance 4000" are 10 displays or printed materials of 100 dpi, 170 dpi, 200 dpi, 300 dpi, and 400 dpi, respectively. The viewing distance when viewed from an inch away.
[0099]
Using the numerical values shown in FIGS. 8 to 10, the square root (W_{b [i]} ^csf・ G_{b [i]})^1/2Was calculated. The calculation results are shown in Tables 7 to 18 below. Tables 7 to 9 are calculated using the numerical values shown in FIG. 8 (9, 7), and black and white numerical values of the filter. Tables 10 to 12 are calculated using the numerical values shown in FIGS. 9 and 10 (9 7) Filter color values, Tables 13 to 15 are calculated using the values shown in FIG. 8, (5, 3) Filter black and white values, Tables 16 to 18 are FIG. 9 and FIG. The numerical values for the color of the (5, 3) filter calculated using the numerical values shown in FIG.
[0100]
[Table 7]

[0101]
[Table 8]

[0102]
[Table 9]

[0103]
[Table 10]

[0104]
[Table 11]

[0105]
[Table 12]

[0106]
[Table 13]

[0107]
[Table 14]

[0108]
[Table 15]

[0109]
[Table 16]

[0110]
[Table 17]

[0111]
[Table 18]

[0112]
Next, using the numerical values shown in Tables 7 to 18, the priority of each band component was calculated in the same procedure as described in the second embodiment. That is, the numerical value of the lowest band component LLn of the highest decomposition level n is α, the numerical values of the other band components are x, the function relating to the variable y rounded to the power of 2 is R [y], and the variable y of 2 Width power 2^mA function for calculating an index m of m = I [2^m] And the absolute value for the variable y is | y |, the priority p is calculated according to p = | I [R [x / α]] |.
[0113]
The priority values are shown in Table 19 to Table 30 below. The priorities of Table 19, Table 20, Table 21, Table 22, Table 23, Table 24, Table 25, Table 26, Table 27, Table 28, Table 29, and Table 30 are shown in Tables 7, 8, and 8, respectively. Table 9, Table 10, Table 11, Table 12, Table 13, Table 14, Table 15, Table 16, Table 17, and Table 18 are used for calculation.
[0114]
[Table 19]

[0115]
[Table 20]

[0116]
[Table 21]

[0117]
[Table 22]

[0118]
[Table 23]

[0119]
[Table 24]

[0120]
[Table 25]

[0121]
[Table 26]

[0122]
[Table 27]

[0123]
[Table 28]

[0124]
[Table 29]

[0125]
[Table 30]

[0126]
In the present embodiment, as in the second embodiment, the priority is set for the conversion coefficient of each band component by shifting right by the number of bits of the priority shown in Tables 19 to 30 above. The Thereby, the priority which considered the human visual characteristic can be set.
[0127]
Image quality control processing.
Next, the configuration and processing contents of the image quality control unit 23 shown in FIG. 1 will be described. FIG. 11 is a functional block diagram showing a schematic configuration of the image quality control unit 23.
[0128]
The image quality control unit 23 selects an image quality parameter QP suitable for the target image quality information from a plurality of image quality parameter groups based on target image quality information (high image quality, standard image quality, low image quality, resolution information, etc.) supplied from the outside. An image quality parameter selection unit 31 that selects and outputs, and an encoding target determination unit 30 that determines an encoding target are provided. The encoding target determination unit 30 sets the above-described priority for each band component of the encoded data BD according to the priority data PD acquired from the priority table 24. The encoding target determination unit 30 determines an encoding target in accordance with the target image quality specified by the image quality parameter QP according to the set priority, and generates and outputs an image quality control signal CS1.
[0129]
Hereinafter, a method for determining an encoding target will be described. FIG. 12 is a schematic diagram illustrating conversion coefficients 33, 33,... Bit-shifted according to priority. Each conversion coefficient 33 is bit-shifted according to the priority. The numbers 0, 1,..., 10 attached to the respective bits of the conversion coefficient 33 indicate the numbers of bit planes to which the bits belong. Here, LSB number = 0 and MSB number = 10.
[0130]
The encoding target determination unit 30 sets the encoding end line 32 according to the image quality parameter QP, determines the higher bits than the encoding end line 32 as the encoding target, and sets the lower bits than the line 32 as the encoding target. The image quality control signal CS1 is generated so as to be excluded from the above. This makes it possible to efficiently select the encoding target. As a result, the arithmetic encoding unit 21 that has received the image quality control signal CS1 arithmetically encodes only the bit plane higher than the encoding end line 32 in each code block, and truncates the bit plane lower than the line 32. It will be. Note that the arithmetic coding unit 21 does not perform arithmetic coding on the bits in which zeros are inserted by the bit shift process.
[0131]
The encoding target determination unit 30 can further determine the encoding target in units of encoding passes according to the image quality parameter QP. The image quality parameter QP includes a parameter group indicating a restriction on a bit plane to be encoded and a restriction on a coding path (CL path, SIG path, and MR path) to be encoded. Table 31 below illustrates image quality parameters QP suitable for an image having a resolution of 2048 × 2560 pixels. Since the resolution of the lowest subband needs to be smaller than 128 × 128 pixels, a resolution level of 5 or higher is required.
[0132]
[Table 31]

[0133]
In Table 31, “Number of bit planes” indicates the number of bit planes subject to truncation of lower bits than the encoding end line 32 shown in FIG. 12, and “Path name” indicates the final encoding pass in the encoding target. The “maximum number of passes” represents the upper limit of the number of encoding passes to be encoded.
[0134]
An example of processing when FIG. 12 and Table 31 are applied will be described below. FIG. 13 shows “00011010111” as the conversion coefficient 33 of the band component LL5.₂= 215_TenExemplify (Y₂Is the binary value Y, X_TenRepresents the decimal value X). As shown in Table 31, the final coding pass in the band component LL5 is limited to the CL pass, and the maximum number of passes is limited to 17.
[0135]
The context determination is made so that the seventh bit of the transform coefficient shown in FIG. 13 belongs to the SIG path or the CL path. The 8th to 10th upper bits are encoded by a method called a tag tree when they belong to a bit plane composed of only 0 bits, and the SIG path when the encoding pass has already started. Alternatively, it is encoded by the CL pass. When the seventh bit belongs to the encoding start pass (CL pass), the lower-order bits including the sixth bit are context-determined so as to belong to the MR pass. In general, the bit planes lower than the encoding start bit plane are encoded in the order of the SIG pass, the MR pass, and the CL pass from the viewpoint of encoding efficiency. Therefore, since the maximum number of passes is limited to 17, a total of 17 passes from the 7th bit CL pass to the 1st bit SIG pass are to be encoded. However, since the first bit belongs to the MR path, it is not encoded. Therefore, in the arithmetic encoding process, the lower 2 bits are discarded, and the encoded value is “00011010100.₂= 212_Ten"If this value is dequantized at midpoint," 00011010110₂= 214_Ten"Become.
[0136]
Next, FIG. 14 shows “00000001111” as the conversion coefficient 33 of the band component LL5.₂= 15_Ten"The third bit of the transform coefficient belongs to the SIG path or the CL path. If the fourth to tenth upper bits belong to a bit plane consisting of only 0 bits, a tag tree (Tag tree)" If the third bit belongs to the encoding start path (CL path), the lower bits including the second bit are encoded. Belongs to the MR pass, and a total of 10 passes from the CL pass of the 3rd bit to the CL pass of the 0th bit are to be encoded.In the arithmetic encoding process, the value after encoding is “00000001111”.₂= 15_Ten"If this value is dequantized," 00000001111₂= 15_Ten"Become.
[0137]
Next, FIG. 15 shows “00001011111” as the conversion coefficient 33 of the band component HH2.₂= 95_Ten“As an example, as shown in Table 31, the final coding pass in the band component HH2 is limited to the SIG pass, and the maximum number of passes is 14. The bit planes of the lower 3 bits are discarded.
[0138]
The sixth bit of the transform coefficient belongs to the SIG path or CL path. The 7th to 10th upper bits are encoded with a tag tree if they belong to a bit plane consisting of only 0 bits, and the SIG pass or CL pass if the encoding pass has already started. It is encoded with. When the sixth bit belongs to the encoding start pass (CL pass), the lower bits including the fifth bit belong to the MR pass. Also, because of the limitation that encoding is performed only up to the SIG pass of the 3rd bit plane, 8 passes from the CL pass of the 6th bit to the SIG pass of the 4th bit are to be encoded, but the 3rd bit is the MR pass. It is not encoded because it belongs to. Therefore, the value after encoding in the arithmetic encoding process is “00001010000₂= 80_Ten"If this value is dequantized at midpoint," 00001011000₂= 88_Ten"Become.
[0139]
The reason why each bit plane is encoded in the order of the SIG pass, MR pass, and CL pass is that the encoding efficiency with respect to the distortion of the SIG pass is the highest. FIG. 16 shows rate / distortion characteristics in each coding pass. Point P in the RD curve₁~ P₂Is the SIG path, point P₂~ P_ThreeIs the MR path, point P_Three~ P_FourIndicates the CL path. Ratio of distortion ΔD to rate (code amount) in each coding pass_SIG/ ΔR_SIG, ΔD_MR/ ΔR_MR, ΔD_CL/ ΔR_CLAs can be seen, the curve slope in the SIG path is the steepest and the coding efficiency is the highest.
[0140]
As described above, in the image quality control method according to the present embodiment, it is determined whether or not the transform coefficient is to be encoded with respect to the transform coefficient bit-shifted according to the priority. Since only the encoding target is selectively arithmetically encoded by the arithmetic encoding unit 21, it is possible to efficiently control the amount of code so that a high-quality compressed image with little distortion can be generated.
[0141]
Code amount control processing.
Next, the configuration and processing contents of the code amount control unit 22 shown in FIG. 1 will be described. FIG. 17 is a functional block diagram showing a schematic configuration of the code amount control unit 22.
[0142]
The code amount control unit 22 includes a mass storage device 40, a code amount calculation unit 41, and a data output control unit 42.
[0143]
As described above, the arithmetic encoding unit 21 shown in FIG. 1 selectively arithmetically encodes only the encoding target designated by the image quality control unit 23, and the encoded data AD obtained as a result is encoded amount control unit 22 Output to. The code amount calculation unit 41 calculates a subtotal of the capacity of the encoded data AD to be input in band component units, bit plane units, and encoding pass units, and outputs the subtotal information 43 obtained as a result to the data output control unit 42 To do. The encoded data AD is temporarily stored in the mass storage device 40 in units of frames or subframes.
[0144]
The data output control unit 42 reads the encoded data AD temporarily stored in the large-capacity storage device 40, and performs bit shift using the priority data PD2 as in the first to third embodiments. Next, the data output control unit 42 calculates a truncation point so as to match the target code amount from a code string generated by rearranging the bit-shifted encoded data in the scanning order described below. Next, the data output control unit 42 outputs the code string before the cut-off point in the code string to the bit stream generation unit 17 as the encoded data CD.
[0145]
18 and 19 are diagrams for explaining an example of the scanning order and the cut-off points. 18 and FIG. 19 display transform coefficients 33, 33,... Bit-shifted according to the priority according to the same rule as shown in FIG. Encoded data AD of higher bits (bits on the left side of the drawing) than the encoding end line 32 is input to the code amount control unit 22.
[0146]
As shown by the arrows in FIG. 18, the transform coefficients 33, 33,... Are in bit-plane units or coding pass units in the order of priority (from the upper bit to the lower bit) and at the same priority. Rearranged in the scanning order from the high frequency side to the low frequency side. In general, as the lower bit plane is encoded, the MR path ratio increases and the compression efficiency tends to decrease. Therefore, in order to encode as many SIG passes as possible in order to improve the compression efficiency, the scanning order from the high frequency side to the low frequency side is adopted at the same priority.
[0147]
Then, the data output control unit 42 determines the truncation point so as to satisfy the condition that the actual code amount (number of bytes) is equal to or less than the target code amount (number of bytes) as shown in the following equation (8), The lower bit plane included in the code string after the truncation point is truncated. As a result, rate control of encoded data that has already been arithmetically encoded can be efficiently performed according to the priority set for each subband.
[0148]
[Equation 8]

[0149]
As shown in FIG. 19, when the second bit plane of the subband HL3 is determined as the truncation point in accordance with the target code amount, the bit indicated by the arrow is truncated.
[0150]
20 is a diagram illustrating a code string rearranged in units of bit planes, and FIG. 21 is a diagram illustrating a code string rearranged in units of coding passes. In FIG. 20, codes LL5, HL5,... Indicating subbands and bit plane numbers 10, 9,. The bit plane after the line 44 attached to the second bit plane of the subband HL3 is discarded.
[0151]
21, for each coding pass, codes CL, SIG, MR indicating the type of coding pass, codes LL5, HL5,... Indicating subbands, bit plane numbers 10, 9,. Is attached. The bit planes after the line 44 attached to the MR path of the second bit plane of the subband HL3 are discarded.
[0152]
As described above, according to the code amount control process according to the present embodiment, it is not necessary to calculate the distortion amount in each encoding pass for the rate / distortion optimization process, so that the real-time property is high and the overhead is greatly increased. Reduced and highly efficient rate control can be realized.
[0153]
【The invention's effect】
As described above, according to the compression coding apparatus according to claim 1 of the present invention, the compression coding method according to claim 9, and the program according to claim 17, the band components are recursively converted into low frequency components, respectively. The priority is set according to the number of times of band division, and it is determined whether or not to be an encoding target according to this priority. Since only the encoding target is selectively entropy-encoded, the code amount can be controlled efficiently, and high-speed encoding processing can be performed with a small amount of computation.
[0154]
According to the second, tenth, and eighteenth aspects, it is possible to generate a compressed image having high display image quality that is suitable for human visual evaluation.
[0155]
According to claim 3, claim 11 and claim 19, the conversion coefficient of each band component is shifted by the number of bits corresponding to the priority, and the priority of the conversion coefficient is determined according to the number of bits to be shifted. It becomes possible to efficiently specify the encoding target in accordance with the image quality of the compressed image.
[0156]
According to the fourth, fifth, sixth, twelfth, thirteenth and fourteenth aspects and the twentieth, twenty-two and twenty-second embodiments, the encoding target can be finely specified according to the priority in bit plane units and encoding pass units. Therefore, it is possible to accurately control the code amount and finely control the image quality of the compressed image.
[0157]
According to the seventh, eighth, fifteenth and sixteenth and twenty-third and twenty-fourth aspects, it is possible to efficiently perform the rate control of the already entropy-coded data according to the priority set for each band component. is there. Further, it is possible to perform rate control with high real-time characteristics so that distortion can be suppressed without performing optimization processing using rate / distortion optimization as in the past.
[Brief description of the drawings]
FIG. 1 is a functional block diagram showing a schematic configuration of a compression encoding apparatus according to an embodiment of the present invention.
FIG. 2 is a schematic diagram showing a two-dimensional image obtained by band division by wavelet transform.
FIG. 3 is a diagram for explaining priority setting processing by bit shift;
FIG. 4 is a diagram illustrating a bit-shifted transform coefficient.
FIG. 5 is a schematic diagram showing a two-dimensional image obtained by band division by wavelet transform.
FIG. 6 is a schematic diagram showing a two-dimensional image obtained by band division by wavelet transform.
7 is a schematic diagram showing band component conversion coefficients that are right-bit shifted in accordance with the priorities shown in FIG. 6;
FIG. 8 is a diagram showing a numerical table of energy weighting factors.
FIG. 9 is a diagram illustrating a numerical table of energy weighting factors.
FIG. 10 is a diagram showing a numerical table of energy weighting factors.
FIG. 11 is a functional block diagram showing a schematic configuration of an image quality control unit according to the present embodiment.
FIG. 12 is a schematic view illustrating transform coefficients bit-shifted according to priority.
FIG. 13 is a diagram for describing an example of encoding processing of a transform coefficient of band component LL5.
FIG. 14 is a diagram for explaining an example of encoding processing of a transform coefficient of band component LL5.
FIG. 15 is a diagram for describing an example of encoding processing of transform coefficients of band component HH2.
FIG. 16 is a diagram showing a curve of rate / distortion characteristics.
FIG. 17 is a functional block diagram illustrating a schematic configuration of a code amount control unit according to the present embodiment.
FIG. 18 is a diagram for explaining an example of a scanning order.
FIG. 19 is a diagram for explaining an example of truncation points;
FIG. 20 is a diagram illustrating code strings rearranged in units of bit planes.
FIG. 21 is a diagram illustrating code strings that are rearranged in units of coding passes.
FIG. 22 is a functional block diagram showing a schematic configuration of a compression encoding apparatus according to the JPEG2000 system.
FIG. 23 is a schematic diagram showing a two-dimensional image that is band-divided according to the octave division method.
FIG. 24 is a schematic diagram showing a two-dimensional image decomposed into a plurality of code blocks.
FIG. 25 is a schematic diagram showing a plurality of bit planes constituting a code block.
FIG. 26 is a schematic diagram showing three types of encoding passes.
FIG. 27 is a diagram showing an RD curve representing the relationship between rate and distortion.
[Explanation of symbols]
1 Compression encoder
10 DC level shift unit
11 Color space converter
12 Tiling Club
13 DWT section
14 Quantizer
15 ROI section
17 Bitstream generator
20 Coefficient bit modeling section
21 Arithmetic coding part
22 Code amount control unit
23 Image quality controller
24 Priority table

Claims (24)

A compression encoding device for compressing and encoding an image signal,
A wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components;
An entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;
Among the plurality of band components, the priority of each band component is set based on the norm of the two-dimensional synthesis filter coefficient of the predetermined band component and according to the number of times the band is recursively divided, An image quality control unit that determines the encoding target to be specified in the entropy encoding unit according to the priority;
A compression encoding apparatus comprising: A compression encoding device for compressing and encoding an image signal ,
A wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components;
An entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;
Priorities are set according to the number of times the band components are recursively band-divided into the low-frequency components, and the encoding target to be specified in the entropy encoding unit is determined according to the priorities. An image quality control unit,
With
The compression coding apparatus, wherein the image quality control unit has a function of using the priority weighted in consideration of human visual characteristics. The compression encoding device according to claim 1 or 2, wherein
The image quality control unit is a compression encoding device that determines the encoding target from a conversion coefficient obtained by shifting the conversion coefficient of each band component by the number of bits corresponding to the priority. Claims 1 a compression encoding apparatus according to any one of claims 3,
The transform coefficient output from the wavelet transform unit is decomposed into a plurality of bit planes configured by two-dimensionally arranging each bit of the transform coefficient, and the bit plane is determined according to the significance of each bit. A coefficient bit modeling unit that generates encoded data by decomposing into determined encoding passes ,
Further comprising
The entropy encoding unit is a compression encoding apparatus that selectively entropy encodes only the designated encoding target among the encoded data generated by the coefficient bit modeling unit. The compression encoding apparatus according to claim 4, wherein
The image quality control unit is a compression encoding device having a function of determining the encoding target in units of the bit planes. The compression encoding apparatus according to claim 4, wherein
The image quality control unit is a compression encoding device having a function of determining the encoding target in units of the encoding pass. We claim 4 a compression encoding apparatus according to any one of claims 6,
A code amount control unit for controlling a rate of encoded data output from the entropy encoding unit ;
Further comprising
The code amount control unit includes:
A code amount calculator that calculates a subtotal of the capacity of encoded data output from the entropy encoder in at least one unit of the band component, the bit plane, and the encoding path;
Using the subtotal calculated by the code amount calculation unit, a truncation point that matches the target code amount is calculated from a code string that is generated by rearranging the encoded data in a predetermined scanning order, and before the truncation point. A data output control unit for outputting the code string of
A compression encoding apparatus comprising: The compression encoding apparatus according to claim 7, wherein
The data output control unit generates the code string by rearranging the transform coefficients in the order of higher priority and in the same priority, in the scanning order from the high frequency side to the low frequency side, Compression encoding device. A compression encoding method for compressing and encoding an image signal,
(A) a step of recursively dividing the image signal into a high-frequency component and a low-frequency component by wavelet transform to generate conversion coefficients of a plurality of band components;
(B) a step of selectively entropy encoding only the specified encoding target among the transform coefficients;
(C) Among the plurality of band components , the priority is set based on the norm of the two-dimensional synthesis filter coefficient of a predetermined band component and according to the number of times the band is recursively divided , and the step Determining the encoding target specified in (b) according to the priority;
A compression encoding method comprising: A compression encoding method for compressing and encoding an image signal ,
(A) a step of recursively dividing the image signal into a high-frequency component and a low-frequency component by wavelet transform to generate conversion coefficients of a plurality of band components;
(B) a step of selectively entropy encoding only the specified encoding target among the transform coefficients;
(C) A priority is set for each band component according to the number of times the band is recursively divided into the low frequency components, and the encoding target specified in the step (b) is set to the priority. A process to be determined according to
With
The compression encoding method, wherein the step (c) includes the step of determining the priority weighted in consideration of human visual characteristics. A compression encoding method according to claim 9 or 10, wherein
The compression encoding method, wherein the step (c) is a step of determining the encoding target from a conversion coefficient obtained by shifting the conversion coefficient of each band component by the number of bits corresponding to the priority. It claims 9 A compression coding method according to any one of claims 11,
After the step (a) and before the step (b),
(D) The transform coefficient generated in the step (a) is decomposed into a plurality of bit planes configured by two-dimensionally arranging each bit of the transform coefficient, and the bit plane is significant for each bit. A process of generating encoded data by decomposing into an encoding pass determined according to the characteristics;
Further comprising
The compression encoding method, wherein the step (b) is a step of selectively entropy encoding only the designated encoding target among the encoded data generated in the step (d). A compression encoding method according to claim 12, comprising:
The step (c) includes a step of determining the encoding target in units of the bit planes. A compression encoding method according to claim 12, comprising:
The step (c) includes a step of determining the encoding target in units of the encoding pass. It claims 12 a compression coding method according to any one of claims 14,
(E) further comprising the step of controlling the rate of the encoded data entropy encoded in the step (b),
The step (e)
(E-1) calculating a subtotal of the capacity of the encoded data entropy encoded in the step (b) in at least one unit of the band component, the bit plane, and the encoding path;
(E-2) Using the subtotal calculated in the step (e-1), a truncation point that matches the target code amount is calculated from a code string generated by rearranging the encoded data in a predetermined scanning order. Outputting the code string before the truncation point;
A compression encoding method comprising: A compression encoding method according to claim 15, comprising:
The step (e-2) generates the code string by rearranging the transform coefficients in the order of high priority and in the same order of priority in the scanning order from the high frequency side to the low frequency side. A compression encoding method including the step of: A program for causing a microprocessor to compress and encode an image signal,
A wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components;
An entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;
Among the plurality of band components , a priority is set based on the norm of a two-dimensional synthesis filter coefficient of a predetermined band component, and according to the number of times of band division recursively , and the entropy coding unit As an image quality control unit that determines the encoding target to be specified according to the priority,
A program for causing the microprocessor to function. A program for causing a microprocessor to compress and encode an image signal ,
A wavelet transform unit that recursively divides an image signal into a high-frequency component and a low-frequency component by wavelet transform to generate and output transform coefficients of a plurality of band components;
An entropy encoding unit that selectively entropy-encodes only a specified encoding target among the transform coefficients;
Priorities are set according to the number of times the band components are recursively band-divided into the low-frequency components, and the encoding target to be specified in the entropy encoding unit is determined according to the priorities. As an image quality control unit
Allowing the microprocessor to function;
The image quality control unit, a program in which said to function the microprocessor to use the priority that has been weighted in consideration of human visual characteristics. A program according to claim 17 or claim 18, wherein
The image quality control unit causes the microprocessor to function so as to determine the encoding target from conversion coefficients obtained by shifting the conversion coefficients of the band components by the number of bits corresponding to the priority. The preceding claims 17 The program according to any one of claims 19,
The transform coefficient output from the wavelet transform unit is decomposed into a plurality of bit planes configured by two-dimensionally arranging each bit of the transform coefficient, and the bit plane is determined according to the significance of each bit. Allowing the microprocessor to function as a coefficient bit modeling unit that decomposes into determined coding passes to generate encoded data;
The entropy encoding unit is a program that causes the microprocessor to selectively entropy encode only the specified encoding target among the encoded data generated by the coefficient bit modeling unit. The program according to claim 20, wherein
The image quality control unit is a program that causes the microprocessor to function so as to determine the encoding target in units of the bit plane. The program according to claim 20, wherein
The image quality control unit is a program that causes the microprocessor to function so as to determine the encoding target in units of the encoding pass. It Claim 20 The method as defined in any one of claims 22,
While causing the microprocessor to function as a code amount control unit that controls the rate of encoded data output from the entropy encoding unit,
The code amount control unit includes:
A code amount calculator that calculates a subtotal of the capacity of encoded data output from the entropy encoder in at least one unit of the band component, the bit plane, and the encoding path;
Using the subtotal calculated by the code amount calculation unit, a truncation point that matches the target code amount is calculated from a code string that is generated by rearranging the encoded data in a predetermined scanning order, and before the truncation point. As a data output control unit for outputting the code string of
A program for causing the microprocessor to function. A program according to claim 23, wherein
The data output control unit generates the code string by rearranging the transform coefficients in the order of high priority and in the scanning order from the high frequency side to the low frequency side in the same priority. A program for causing the microprocessor to function.

Images