JP3759932B2 - Image compression method and decompression method - Google Patents

Description

分散値Ｖｍｎは、画像信号のもつ高周波成分の量に応じて大きくなる。この分散値Ｖｍｎが閾値Ｔｈよりも小さい場合は低画質で良い領域、閾値Ｔｈよりも大きい場合は高画質が要求される領域と判断される。なお、閾値Ｔｈは、所望の圧縮率、量子化テーブルの値等との関係から決定する。また、閾値Ｔｈを複数設定して、閾値Ｔｈ毎に画質を決定すると、復元された画像の画質を領域毎にきめ細やかに変えることができる。
【００２７】
ここで、分割する領域の画素の所定数は画像全体において一定であっても良いし、ある領域は３２×３２画素に、別の領域は１６×１６画素に分割する等にして分割する画素数を領域毎に変えても良い。また、後述のダウンサンプリングにおいても、任意の画素数でダウンサンプリングして良い。また、領域を分割する場合と同様に、圧縮・伸張方法としてＪＰＥＧを使用する場合には、８×８画素、１６×１６画素のように８×８画素の整数倍でダウンサンプリングすると、ＪＰＥＧの圧縮・伸張処理の単位である８×８画素の倍数となり処理が容易となる。
［ダウンサンプリング部］
例えば、画像を所定数の画素＝１６×１６画素に分割し、低画質で良い領域のダウンサンプリングに８×８画素を適用した場合、
分割部１１０で１６×１６画素の領域に分割され、ダウンサンプリング部１２０により低画質で良い領域は、１６×１６画素から８×８画素にダウンサンプリングされる。一方、高画質が要求される領域は、１６×１６画素のままでダウンサンプリングされない。また、ダウンサンプリング部１２０は、ダウンサンプリングされている領域とダウンサンプリングされない領域とを合算して符号化器１３０に出力する。
【００２８】
ダウンサンプリングは、領域内の画素を抜き出す間引き処理やフィルタ処理により行われている。フィルタ処理としては、ローパスフィルタ処理やガウシアンフィルタ処理等がある。間引き処理のみでダウンサンプリングを行っても良い。また、フィルタ処理と間引き処理とを併用してダウンサンプリングを行うと歪みの少ない精度の良いダウンサンプリングが可能である。
例えば、ガウシアンフィルタ処理と間引き処理とによってダウンサンプリングが行われる場合には、次のように処理が行われる。ガウシアンフィルタ処理は、一次元ガウシアンフィルタを画像信号の水平方向に適用し、その結果得られる信号に対して垂直方向に同じフィルタを適用して行われる。また、ガウシアンフィルタ処理の後に画素の間引きを行うことによって所望のサイズにダウンサンプリングされた画像が得られる。
【００２９】
以下に、１／２のダウンサンプリングの一例を図２（ａ）を参照して説明する。図２（ａ）は、１６×１６画素の１領域を示している。ダウンサンプリングの方法としては、１６×１６画素のデータのダウンサンプリングした画像（以下、代表値という）を８×８画素のδｓ部分（Ｘ＝０〜７、Ｙ＝０〜７）に格納し、その他のαｓ（Ｘ＝８〜１５、Ｙ＝０〜７）、βｓ（Ｘ＝０〜７、Ｙ＝８〜１５）、γｓ（Ｘ＝８〜１５、Ｙ＝８〜１５）領域にフィルビットとして“０”を格納する方法がある。フィルビットとしては、特に“０”を使用する必要はなく、別の同じ数値を用いることもできる。例えば、１６×１６画素のデータの代表値を８×８画素のδｓ部分（Ｘ＝０〜７、Ｙ＝０〜７）に格納し、その他のαｓ（Ｘ＝８〜１５、Ｙ＝０〜７）、βｓ（Ｘ＝０〜７、Ｙ＝８〜１５）、γｓ（Ｘ＝８〜１５、Ｙ＝８〜１５）領域にフィルビットとして１６×１６画素の平均値、最大値、最小値などの同一の数値を挿入して行う。
【００３０】
このように、ダウンサンプリングして残りの画素に“０”または同じ数値を挿入することで、符号化の際のエントロピー符号化により効率的に画像を圧縮することができ、圧縮率を高めることができる。また、フィルビットを使用することによりダウンサンプリングされたことを示すフラグ等は不要で、圧縮率を低下させることもない。
閾値Ｔｈ毎にダウンサンプリングする画素数を決定すると、復元された画像の画質を領域毎にきめ細やかに変えることができる。例えば、ある領域は３２×３２画素から１６×１６画素の１／２に、別の領域は３２×３２画素から８×８画素に１／４のダウンサンプリングする等してダウンサンプリングする画素数を変えても良い。また、ある領域は３２×３２画素から８×８画素の１／４に、別の領域は１６×１６画素から８×８画素の１／２のダウンサンプリングする等、分割する画素数を変えてダウンサンプリングを行っても良い。このように分割する画素数とダウンサンプリングする画素数とをそれぞれ変えたり、あるいは組み合わせて変えることで、１／２，１／４，１／３，２／３等様々なダウンサンプリングを適用することで復元された画像の画質を領域毎にきめ細やかに変えることができる。
［符号化器、Ｉ／Ｆ］
符号化器１３０は、単一の量子化テーブルを用いる汎用の圧縮方法で圧縮を行い、Ｉ／Ｆ１４０を経由して伸張側に圧縮データが伝送される。単一の量子化テーブルを用いる汎用の圧縮方法は、例えばＪＰＥＧが挙げられる。
【００３１】
次に、伸張側のコンピュータ２００の機能構成の説明を行う。
［復号化器、Ｉ／Ｆ］
圧縮データは、Ｉ／Ｆ１５０を経由して復号化器１６０に入力され、単一の量子化テーブルを用いる汎用の伸張方法で伸張される。伸張は、圧縮方法と対応した方法により行われ、例えば、圧縮がＪＰＥＧにより行われている場合には、ＪＰＥＧにより伸張が行われる。
［判定部］
判定部１７０は、ダウンサンプリングされている領域を判定する。ダウンサンプリングされている領域を判定する方法としては、フィルビットを検出する方法がある。
【００３２】
例えば、分割には１６×１６画素を、ダウンサンプリングには８×８画素を、フィルビットには“０”を適用し、前記図２（ａ）においてダウンサンプリングされた場合、αｓ、βｓ、γｓ領域にはフィルビットとして全て“０”が格納される。よって、αｓ、βｓ、γｓ領域全てに“０”が格納されている場合にはダウンサンプリングがされていることを、全てが“０”ではない場合はダウンサンプリングされていないことを判定することができる。αｓ、βｓ、γｓ領域全てに“０”が格納されているかどうかは、図２（ａ）に基づいた下記式（２）により表される。
【００３３】
【数２】

ここで、αｓ、βｓ、γｓ領域に全て“０”が格納されている場合には、Ｄｍｎは“０”となりダウンサンプリングされたことを検出することができる。
また、分割には３２×３２画素を、ダウンサンプリングには領域毎の画質に対応して１６×１６画素または８×８画素を、フィルビットには“０”を適用してダウンサンプリングする画素数を変えた場合のダウンサンプリングの判定方法を、図２（ａ）及び図２（ｂ）を参照して説明する。図２（ｂ）は、３２×３２画素の１領域を示している。まず、図２（ｂ）のように３２×３２画素の領域を４分割し、αＬ（Ｘ＝１６〜３１、Ｙ＝０〜１５）、βＬ（Ｘ＝０〜１５、Ｙ＝１６〜３１）、γＬ（Ｘ＝１６〜３１、Ｙ＝１６〜３１）の全てに“０”が格納されているかどうかを前記式（２）により判定する。次に、αＬ、βＬ、γＬが全て“０”である場合は、図２（ａ）において、αｓ（Ｘ＝８〜１５、Ｙ＝０〜７）、βｓ（Ｘ＝０〜７、Ｙ＝８〜１５）、γｓ（Ｘ＝８〜１５、Ｙ＝８〜１５）の全てに“０”が格納されているかどうかを、同様に判定する。このような判定により、（１）αＬ、βＬ、γＬに“０”が格納されていない場合は、ダウンサンプリングされていない。（２）αＬ、βＬ、γＬに“０”が格納され、かつαｓ、βｓ、γｓに“０”が格納されていない場合は、３２×３２画素から１６×１６画素にダウンサンプリングされている。（３）αＬ、βＬ、γＬに“０”が格納され、かつαｓ、βｓ、γｓに“０”が格納されている場合は、３２×３２画素から８×８画素にダウンサンプリングされていると判定することができる。
【００３４】
さらに、分割には領域毎の画質に対応して３２×３２画素または１６×１６画素を、ダウンサンプリングには８×８画素を、フィルビットには“０”を適用した場合のダウンサンプリングの判定方法を、図２（ａ）、（ｂ）を参照して説明する。まず、図２（ａ）において、１６×１６画素のうちαｓ、βｓ、γｓの全てに“０”が格納されているかどうかを、前記式（２）により判定する。次に、αｓ、βｓ、γｓが全て“０”である場合は、このαｓ、βｓ、γｓを含む１６×１６画素を、図２（ｂ）のδＬとして、αＬ、βＬ、γＬの全てに“０”が格納されているかどうかを同様に判定する。このような判定により、（１）αｓ、βｓ、γｓに“０”が格納されていない場合は、ダウンサンプリングされていない。（２）αｓ、βｓ、γｓに“０”が格納され、かつαＬ、βＬ、γＬに“０”が格納されていない場合は、１６×１６画素から８×８画素にダウンサンプリングされている。（３）αｓ、βｓ、γｓに“０”が格納され、かつαＬ、βＬ、γＬに“０”が格納されている場合は、３２×３２画素から８×８画素にダウンサンプリングされていると判定することができる。
【００３５】
この他、ダウンサンプリングを判定する方法は種々考えられる。また、αｓ、βｓ、γｓ領域に全て“０”が格納されていることを検出するのではなく、別の同じ数値で全て満たされていることを検出することによりサンプリング方法を検出してもよい。
このように、αｓ、βｓ、γｓ領域全てに“０”あるいはその他の同じ数値が格納されていることを検出することで、ダウンサンプリングされている領域を判定することができる。
［補間部、出力部］
判定部１７０の判定結果により、補間部１８０は、ダウンサンプリングされている領域を８×８画素から１６×１６画素等、元の画素数に補間する。一方、ダウンサンプリングされていない領域は補間を必要としない。また、補間部１８０は、ダウンサンプリング後補間された領域とダウンサンプリングされず補間されていない領域とを合算して出力部１９０に出力する。出力部１９０は、例えばＬＣＤ（図示しない）の表示画素数にあわせて表示できるように、画像データを変換する。よって、ダウンサンプリングされている領域の画素をダウンサンプリング後のデータに基づいて補間して当該領域の画像を出力することができる。また、ダウンサンプリングされていない領域の画像は補間の必要が無いので、ダウンサンプリングされている領域の画像よりも高画質の画像で出力することができる。
【００３６】
補間する方法としては、例えば失われた画像の高周波成分の推定が可能なＬＰ拡大法、線形補間、２次補間、３次補間等を使用することができる。ＬＰ拡大法とは、ダウンサンプリング部１２０によるダウンサンプリングの際に失われた高周波成分を推定して画像の復元を行う方法である。
本発明による圧縮・伸張システムは、複数の領域からなる画像を符号化する前に領域毎に要求される画質に応じてダウンサンプリングを行うので、単一の量子化テーブルにより画像圧縮・伸張をして領域毎に画質を変えることができる。よって、単一の量子化テーブルを用いた汎用フォーマットを用いて、領域毎に画質を変えて画像圧縮・伸張をすることができる。また、領域毎の画質に応じて少なくとも一つの領域のダウンサンプリングを行うことにより、単に圧縮する場合に比べて画像全体の圧縮効率を高めることができ、しかも高画質が必要な領域を高画質で出力することができる。ダウンサンプリングする画素数を領域毎に変えると、画質の変化の自由度を高めることができる。圧縮効率を高めることで、ネットワーク３００による伝送効率を高めることができる。
【００３７】
また、分割部１１０とダウンサンプリング部１２０とを符号化器１３０内に取り込むことができる。同様に、判定部１７０と補間部１８０とを復号化器１６０内に取り込むことができる。このように符号化器１３０、復号化器１６０内に様々な処理を取り込むことで構成が簡単になり、処理速度向上、機器の小型化、低コスト化などを図ることができる。
さらに、図１に示す圧縮・伸張システムは、圧縮側のコンピュータ１００と伸張側のコンピュータ２００とがネットワーク３００で接続されているが、圧縮・伸張を同一のコンピュータによって行い、ネットワーク３００を介さない圧縮・伸張システムであっても良い。このように、ネットワークを介さない場合であっても、画像を保存する際の容量を削減したり、画像の保存枚数を増やすことができる。
＜性能評価＞
次に、実施形態例に係る画像圧縮・伸張システムにより圧縮・伸張した画像の評価を行ったので、その評価について説明する。評価においては、圧縮・伸張方法としてＪＰＥＧを用いている。
【００３８】
画像圧縮・伸張による画像の品質の良否を判断する方法としては例えば以下の方法が挙げられる。その一つは、横軸にファイルサイズ（圧縮率）、縦軸にＰＳＮＲ（Ｐｅａｋ Ｓｉｇｎａｌ ｔｏ Ｎｏｉｓｅ Ｒａｔｉｏ）をとったグラフにより数値評価する方法である。これによって、同じファイルサイズでの画像の品質を評価したり、逆に同じ画像品質でのファイルサイズを数値により評価することができる。ここで、ＰＳＮＲとは符号化による画像の劣化の尺度を表す。つまり、圧縮前の画像と圧縮・伸張後の復元画像との差異を評価する尺度であり、ＰＳＮＲが小さくなるほど画像が劣化する。もう一つの方法は、数値評価により画像の品質を評価するのではなく、主観により評価する方法である。
【００３９】
以下に、ＰＳＮＲを用いた評価方法により画像品質の評価を行った結果と主観により評価を行った結果とを説明する。なお、使用した圧縮・伸張前の原画像は、８ｂｉｔグレースケールでサイズが２５６×２５６の画像である（図６の原画像参照）。
まず、ＰＳＮＲを用いた評価方法による結果を説明する。表１に評価した３種類の実験の内容を示し、以下に表１の３種類の実験の具体的内容を示す。
【００４０】
【表１】

（１）ダウンサンプリング１＋ＪＰＥＧ：原画像を１６×１６画素の領域に分割し、領域の分散値Ｖｍｎが閾値Ｔｈより小さい場合（Ｖｍｎ＜Ｔｈ）は、８×８画素にダウンサンプリングし、分散値Ｖｍｎが閾値Ｔｈより大きい場合（Ｖｍｎ＞Ｔｈ）はダウンサンプリングを行わない。そのダウンサンプリング後の画像データをＪＰＥＧ符号化する。なお、図３は、画像が１６×１６画素に分割され、分割された領域のうち８×８画素にダウンサンプリングされている領域には、フィルビットが挿入されている様子を示している（領域Ａ参照）。また、ダウンサンプリングされていない領域にはフィルビットは挿入されていない（領域Ｂ参照）。
（２）ダウンサンプリング２＋ＪＰＥＧ：原画像を３２×３２画素の領域に分割し、領域の分散値Ｖｍｎが閾値Ｔｈ２より小さい場合（Ｖｍｎ＜Ｔｈ２）は、８×８画素にダウンサンプリングし、分散値Ｖｍｎが閾値Ｔｈ２より大きく、かつ閾値Ｔｈ１より小さい場合（Ｔｈ２＜Ｖｍｎ＜Ｔｈ１）は、１６×１６画素にダウンサンプリングし、分散値Ｖｍｎが閾値Ｔｈ１より大きい場合（Ｖｍｎ＞Ｔｈ１）はダウンサンプリングを行わない。そのダウンサンプリング後の画像データをＪＰＥＧ符号化する。なお、図４は、画像が３２×３２画素に分割され、分割された領域のうち８×８画素または１６×１６画素にダウンサンプリングされている領域には、フィルビットが挿入されている様子を示している（領域ＣまたはＤ参照）。また、ダウンサンプリングされていない領域にはフィルビットは挿入されていない（領域Ｅ参照）。
（３）ＪＰＥＧ：原画像をダウンサンプリングすることなくＪＰＥＧ符号化する。
【００４１】
なお、（３）の実験は、本発明を適用した（１）、（２）の実験と比較するための実験である。
図５は、表１の実験結果をビットレート（圧縮率）に対するＰＳＮＲにより評価をした結果を示す。図５より、ビットレートが０．２〜０．５（ｂｉｔ／ｐｉｘｅｌ）の全ての場合において、本発明を適用した（１）、（２）のＰＳＮＲが、本発明を適用していない（３）のＰＳＮＲよりも大きい。よって、ＰＳＮＲによる画像品質の評価においては、本発明を適用した場合の画像圧縮・伸張は優れていると評価できる。特に、閾値毎にダウンサンプリングの画素数を変えた方法（（２）ダウンサンプリング２＋ＪＰＥＧ）は、ＰＳＮＲが大きく、つまり圧縮前の画像と圧縮・伸張後の復元画像との差異が少ないため優れた画像品質を有していると評価できる。
【００４２】
次に、主観により評価を行うために、図６〜図９に上記の実験による画像圧縮・伸張後の画像を示す。図６〜図９は、それぞれ順に０．５，０．４，０．３，０．２（ｂｉｔ／ｐｉｘｅｌ）の圧縮率が適用された場合の復元画像である。
図６〜図９の全てにおいて、ＪＰＥＧ符号化のみを適用した（３）より本発明の領域毎のダウンサンプリングを適用した（１）、（２）の方がダウンサンプリングされていない領域（図３、図４において領域Ｂまたは領域Ｅ）の画像が鮮明であることが分かる。また、閾値毎にダウンサンプリングの画素数を変えた場合（２）の方が、一定の画素数でダウンサンプリングした場合（１）よりもダウンサンプリングされていない領域の画質が良い。また、画像の圧縮率が上昇するほど、（３）＜（１）＜（２）の順で高画質が要求される領域（領域Ｂまたは領域Ｅ）の画質が良くなることが分かる。
【００４３】
次に、図１０、図１１にダウンサンプリングする領域としない領域とを手動で分割し、ダウンサンプリングの画素数を決定した場合の実験結果を示す。図１０は、人物の部分と背景部分とを手動で分割した後、背景部分をそれぞれ１６×１６画素から８×８画素へダウンサンプリングして圧縮率を０．４（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像とＪＰＥＧ符号化のみにより圧縮率を０．４（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像とを示す。一方、図１１は、背景部分をそれぞれ３２×３２画素から８×８画素へダウンサンプリングして圧縮率を０．３（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像とＪＰＥＧ符号化のみにより圧縮率を０．３（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像とを示す。図１０、図１１の両方において、ＪＰＥＧのみの符号化よりも本発明によるダウンサンプリングを適用して符号化した方が、ダウンサンプリングされていない人物の部分の画像が鮮明であることが分かる。
【００４４】
上記の結果より、機械的、あるいは手動によって本発明の領域毎のダウンサンプリングを行えば、同じ圧縮率であっても必要とする領域の画質をその他の領域の画質より高めることができる。つまり、同じ画質であれば高い圧縮率が得られる。
［本発明のその他への適用］
本発明は、動きの変化のある領域についての画質を変更するのに使用することもできる。例えば、動きの変化のある領域のみ画質をあげてその他の領域の画質を下げることで、動きの変化のある領域の画質を低下させることなく全体として圧縮率を高めることができる。
【００４５】
さらに、本発明では、複数の画像を合わせて一画面に並べて表示するようにした画像に対して、個別の画像領域毎に画質を変更するようにしても良い。例えば、複数の監視カメラにより複数地点の画像を取得し、カメラ毎の画像を前述のように複数並べて一画像とし、監視用モニタに表示する。この画像を圧縮する際、カメラ毎の画像領域に対して画質設定を受け付けることで、カメラ毎の画像の画質を変更することができる。このとき、カメラ近傍のセンサにより、動きの変化のあったカメラの画像の画質を上げ、その他の画像の画質を自動的に下げるようにしても良い。
【００４６】
【発明の効果】
本発明を用いれば、単一の量子化テーブルを用いた汎用フォーマットを用いつつ領域毎の画質を変えることができる。
【図面の簡単な説明】
【図１】本発明の実施形態例に係る画像圧縮・伸張システムの一例。
【図２】（ａ）１６×１６画素の１領域。
（ｂ）３２×３２画素の１領域。
【図３】画像が１６×１６画素に分割され、フィルビットが挿入されている様子。
【図４】画像が３２×３２画素に分割され、フィルビットが挿入されている様子。
【図５】表１の実験結果をビットレートに対するＰＳＮＲにより評価をした結果。
【図６】０．５（ｂｉｔ／ｐｉｘｅｌ）の圧縮率が適用された場合の復元画像。
【図７】０．４（ｂｉｔ／ｐｉｘｅｌ）の圧縮率が適用された場合の復元画像。
【図８】０．３（ｂｉｔ／ｐｉｘｅｌ）の圧縮率が適用された場合の復元画像。
【図９】０．２（ｂｉｔ／ｐｉｘｅｌ）の圧縮率が適用された場合の復元画像。
【図１０】人物の部分と背景部分の内、背景部分をダウンサンプリングして圧縮率を０．４（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像。
【図１１】人物の部分と背景部分の内、背景部分をダウンサンプリングして圧縮率を０．３（ｂｉｔ／ｐｉｘｅｌ）にした場合の画像。
【図１２】圧縮・伸張システムの一例。
【図１３】輝度（Ｙ）・色差（ＵＶ）ごとの量子化テーブル。
【図１４】領域毎に異なる量子化テーブルにより領域毎に画質を変更する場合の圧縮・伸張システムの一例。
【符号の説明】
１：圧縮器
２：伸張器
１０、１０５：入力部
２０：ＤＣＴ変換器
３０：量子化器
３５、６５：量子化テーブル
４０：符号化器
５０：復号化器
６０：逆量子化器
７０：逆ＤＣＴ変換器
８０、１９０：出力部
１００、２００：コンピュータ
１１０：分割部
１２０：ダウンサンプリング部
１３０：符号化器
１４０、１５０：Ｉ／Ｆ
１６０：復号化器
１７０：判定部
１８０：補間部
３００：ネットワーク[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image compression / decompression method and apparatus having different image quality for each region.
[0002]
[Prior art]
As an image compression / decompression method, there is, for example, JPEG (Joint Photographic Graphics) standardized by CCITT (International Telegraph and Teleconductive Committee) and ISO (International Standards Organisation). In the JPEG standard image compression / decompression, 8 × 8 pixels are divided into 1 block (1 unit), an image for one screen is divided into a plurality of blocks, compressed by a JPEG encoder, and decompressed by a JPEG decoder. Yes. The color expression method is performed by a YUV format using luminance (Y) and color difference (UV), and DCT (Discrete Cosine Transform), quantization, encoding, and the like are performed for each YUV.
[0003]
[Problems to be solved by the invention]
FIG. 12 shows an example of a JPEG compression / decompression system. The compression / decompression system shown in FIG. 12 includes a compressor 1 that compresses image data input from the input unit 10, and a decompressor 2 that decompresses the compressed image and outputs a restored image to the output unit 80. ing. The compressor 1 has a DCT converter 20, a quantizer 30 and an encoder 40, and the decompressor 2 has a decoder 50, an inverse quantizer 60 and an inverse DCT converter 70. Yes. The compression / decompression system configured as described above performs image compression / decompression as follows.
[0004]
In the image compression, first, an input image is divided into a plurality of blocks of 8 × 8 pixels, and each block is converted into 8 × 8 DCT coefficients by the DCT converter 20. Next, 8 × 8 DCT coefficients are quantized based on the quantization table 35 of the quantizer 30. The quantization table 35 is, for example, coefficients represented by 8 × 8 for each luminance (Y) / color difference (UV) shown in FIG. Further, the quantized DCT coefficient is subjected to image compression by entropy coding using, for example, Huffman coding or the like by the encoder 40.
[0005]
On the other hand, in the image expansion, first, the compressed image is decoded by the decoder 50, and then inversely quantized based on the quantization table 65 of the inverse quantizer 60. Finally, the inverse DCT converter 70 restores the image and outputs the restored image.
The quantization tables 35 and 65 used for the compression / decompression are commonly used for all blocks (8 × 8 pixels) constituting the image. That is, JPEG does not have different quantization tables 35 and 65 for each block, but has common quantization tables 35 and 65 for all blocks. Therefore, the image cannot be compressed / expanded by changing the image quality of the specific area and the image quality of the other areas.
[0006]
Therefore, as a method of compressing / decompressing different image quality for each region, a method having a quantization table different for each region is used. FIG. 14 shows an example of a compression / decompression system in the case where the image quality is changed for each area by using a quantization table different for each area.
The compression / decompression system shown in FIG. 14 has basically the same configuration as the compression / decompression system shown in FIG. 12 is different from the compression / decompression system shown in FIG. 12 in that the input image is composed of a plurality of regions a1, a2, a3... And the quantizer 30 and the inverse quantizer 60 are provided for each region a1, a2, a3. Have quantization tables a1, a2, a3. The image compression / decompression method is different in that quantization is performed using a different quantization table for each region of the input image, and inverse quantization is performed using a different quantization table for each region. Therefore, the regions a1, a2, a3... Are quantized / inversely quantized according to the quantization tables a1, a2, a3.
[0007]
However, the compression / decompression method using a different quantization table for each area as described above is a compression / decompression method of a unique format, and compression / decompression according to a standard using a single quantization table common to all areas. Not a way. Therefore, a general format using a single quantization table cannot be used, or an image compressed by a compression method using a different quantization table for each region is used as a general format using a single quantization table. Even if decompression is performed by the above method, there arises a problem that the image cannot be restored because the quantization table applied to the quantization is different.
[0008]
SUMMARY OF THE INVENTION An object of the present invention is to provide a compression / decompression method and apparatus capable of changing the image quality for each area while using a general-purpose format using a single quantization table.
[0009]
[Means for Solving the Problems]
In order to solve the above problems, the first invention of the present application is:
A dividing step of dividing the image into a plurality of regions;
-At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least part of the area. A downsampling step of inserting fill bits in other parts of the region;
A quantization step for quantizing the image that has undergone the division step and the downsampling step with a common quantization table for the plurality of regions;
An encoding step for encoding the image quantized in the quantization step;
A decoding step for decoding the image encoded in the encoding step;
An inverse quantization step of inversely quantizing the image decoded in the decoding step with the quantization table;
An interpolation step of interpolating pixels in a region that is down-sampled among a plurality of regions constituting the image dequantized in the inverse quantization step;
An image output step of outputting the image with an image quality required for each of the plurality of regions;
An image compression / decompression method is provided.
[0010]
At least one region of the image is down-sampled according to the image quality required for the region, the region is filled with the representative value and the fill bit, and the entire image is encoded / decoded. Therefore, even if image compression / decompression is performed using a single quantization table, the image quality can be changed for each region. Therefore, image quality can be improved for each region using image compression / decompression in a general-purpose format using a single quantization table. Can be changed. Also, by downsampling at least one area according to the image quality of each area and inserting fill bits, the compression efficiency of the entire image can be increased compared with the case of simply compressing, and the image quality is good. Can output the required area with high image quality.
[0011]
By using the fill bit, a flag or the like indicating the area down-sampled to the decoding side is unnecessary, and there is no need to add redundant data. For example, when “0” is used as the fill bit, the portion into which the fill bit is inserted can be efficiently compressed by entropy coding, so that the compression ratio of the entire image can be increased. The fill bit may not be “0”, but may be other numerical values.
The second invention of the present application is
A dividing step of dividing the image into a plurality of regions;
-At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least part of the area. A downsampling step of inserting fill bits in other parts of the region;
A quantization step for quantizing the image that has undergone the division step and the downsampling step with a common quantization table for the plurality of regions;
An encoding step for encoding the image quantized in the quantization step;
An image compression method is provided.
[0012]
By downsampling for each area, the compression efficiency of the entire image can be increased as compared with the case of simply compressing.
A third invention of the present application provides the image compression method according to the second invention of the present application, wherein the quantization step and the encoding step are performed by JPEG (Joint Photographic Expert Group).
By encoding using JPEG, a general format using a single quantization table generally used in image compression can be applied.
[0013]
According to a fourth invention of the present application, in the second invention of the present application, the dividing step includes a dividing instruction step for receiving an instruction to divide the image into the plurality of areas, and an image quality required for at least one of the plurality of areas. An image compression method including an image quality acceptance step for accepting the setting of the image quality is provided.
Since it is possible to manually specify the area for changing the image quality and set the image quality, the degree of freedom in changing the image quality can be increased.
A fifth invention of the present application is the second invention of the present application, wherein the dividing step includes an image dividing step of dividing the image into the plurality of regions including a predetermined number of pixels, and a signal amount indicating a high-frequency component for each of the plurality of regions. An image compression method including an image quality determination step for determining an image quality required for each region according to the size of the image.
[0014]
Therefore, the image is mechanically divided into regions each having a predetermined number of pixels, and the signal amount for each region is calculated to determine the image quality, so that the image quality can be automatically set. In addition, manpower and costs can be reduced.
The sixth invention of the present application down-samples at least one of a plurality of regions constituting the image and a dividing unit that divides the image into a plurality of regions according to the image quality required for the region, and at least one of the regions. A plurality of the downsampling means for inserting a down-sampled image of the region (hereinafter referred to as a representative value) in a part, a fill-bit in another part of the region, the dividing unit and the downsampling unit There is provided an image compression apparatus including a quantizer that performs quantization using a common quantization table and an encoder that encodes an image quantized by the quantizer.
This device has the same effects as the second invention.
[0015]
The seventh invention of the present application is
A dividing means for dividing the image into a plurality of areas;
-At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least part of the area. Down-sampling means for inserting fill bits in other parts of the region;
A quantization unit that quantizes the image that has passed through the dividing unit and the downsampling unit using a common quantization table for the plurality of regions; and
Encoding means for encoding the image quantized by the quantization means;
An image compression program for causing a computer to function is provided. This program has the same effects as the second invention.
[0016]
The eighth invention of the present application is
Decoding that is divided into a plurality of regions, and after at least one of the plurality of regions is downsampled, the plurality of regions are quantized by a common quantization table and decoded. Steps,
An inverse quantization step of inversely quantizing the image decoded in the decoding step with the quantization table;
An interpolation step of interpolating pixels in a region that is down-sampled among a plurality of regions constituting the image dequantized in the inverse quantization step;
An image output step of outputting the image with an image quality required for each of the plurality of regions;
An image decompression method is provided.
[0017]
Decodes an image that has been divided and encoded into multiple regions, and when a part of the image is downsampled, it interpolates the pixels in that region and outputs the image with the image quality required for each region To do. Therefore, since the image quality can be changed for each area by image expansion using a single quantization table, the image quality can be changed for each area by image expansion in a general-purpose format using a single quantization table. For this reason, it is possible to output a region requiring high quality image quality with high image quality.
A ninth invention of the present application provides the image decompression method according to the eighth invention of the present application, wherein the decoding step and the inverse quantization step are performed by JPEG.
[0018]
By performing decoding and inverse quantization using JPEG, a general-purpose format using a single quantization table generally used in image expansion can be applied, and output image quality can be improved for each region. Can be changed.
The tenth aspect of the present invention provides the image decompression method according to the eighth aspect of the present invention, wherein the interpolation step further includes a step of specifying a downsampled region based on whether or not a predetermined fill bit is inserted. To do.
Therefore, the downsampled area can be specified without adding a flag or the like.
[0019]
The eleventh invention of the present application is
Decoding that is divided into a plurality of regions, and after at least one of the plurality of regions is downsampled, the plurality of regions are quantized by a common quantization table and decoded. And
An inverse quantizer that inversely quantizes the image decoded by the decoder using the quantization table;
Interpolating means for interpolating pixels in a downsampled region among a plurality of regions constituting an image quantized by the inverse quantizer;
Image output means for outputting the image with image quality required for each of the plurality of areas;
An image expansion apparatus including the above is provided. This device has the same effects as the eighth invention.
[0020]
The twelfth invention of the present application
Decoding that is divided into a plurality of regions, and after at least one of the plurality of regions is downsampled, the plurality of regions are quantized by a common quantization table and decoded. means,
An inverse quantization means for inversely quantizing the image decoded by the decoding means using the quantization table;
Interpolating means for interpolating pixels in a downsampled region among a plurality of regions constituting the image dequantized in the dequantizing means; and
Image output means for outputting the image with the image quality required for each of the plurality of areas;
An image expansion program for causing a computer to function is provided. This program has the same effects as the eighth invention.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an image compression / decompression system according to the present invention will be described in detail with reference to an embodiment. FIG. 1 shows an example of an image compression / decompression system according to an embodiment of the present invention.
[overall structure]
The compression / decompression system includes a computer 100 and a computer 200 connected via a network 300. The computer 100 includes an input unit 105, the compressor 1, and an interface (hereinafter referred to as I / F) 140. On the other hand, the computer 200 includes an I / F 150, a decompressor 2, and an output unit 190. The network 300 is constructed by a packet communication network such as Ethernet (registered trademark), the Internet, ATM (Asynchronous Transfer Mode), and other digital networks. The compressor 1 includes a dividing unit 110, a downsampling unit 120, and an encoder 130. The decompressor 2 includes a decoder 160, a determination unit 170, and an interpolation unit 180. In each part of the compression / decompression system having the above-described configuration, image compression / decompression is performed as follows.
[0022]
First, the functional configuration of the computer 100 on the compression side will be described.
[Input section]
The input unit 105 provides image data corresponding to the compressor 1 to the compressor 1. The input unit 105 includes an image capturing part such as a camera or a scanner, a storage part such as an HDD for storing an image, a conversion part for converting image data into image data corresponding to the compressor 1, and the like.
[Division part]
The dividing unit 110 divides the image data captured from the input unit 105 into a plurality of areas a1, a2, a3... And determines the required image quality for each divided area.
[0023]
Here, the required image quality is an image quality in response to a request that a certain area has no problem even if the image quality is lowered, and another certain area does not deteriorate the image quality.
The size of the area to be divided may be an arbitrary number of pixels. When JPEG is used as the compression / decompression method, it is divided into integer multiples of 8 × 8 pixels, such as 8 × 8 pixels, 16 × 16 pixels, and 8 × 8 pixels, which is the unit of JPEG compression / decompression processing. It becomes a multiple of and processing becomes easy.
There are various methods for dividing an input image into a plurality of regions and a method for determining the image quality for each region. As one method, a part for which a reduction in image quality is prevented and a part for which the image quality may be reduced are determined by humans. There is a method of dividing the region by judging and determining with the eyes. For example, it is divided into a region of a target portion where it is desired to prevent a reduction in image quality and a region of a background portion where image quality other than the target portion may be reduced. When the image is manually divided, designation of the area of the divided image and setting of image quality required for the area are accepted. As described above, since the division of the region and the setting of the image quality can be manually performed, the target portion and the background portion can be reliably divided, and the degree of freedom in changing the image quality can be increased.
[0024]
As another method of dividing the input image into a plurality of regions and determining the image quality for each region, the image is mechanically divided into a plurality of regions composed of a predetermined number of pixels, and each divided region is divided. There is a method for mechanically determining the image quality. As a method of mechanically determining the image quality for each region, there is a method of determining by a method using the magnitude of the signal amount indicating the high frequency component of each region, for example, the magnitude of the variance value Vmn. In this way, the image compression processing can be automatically performed by mechanically dividing the image and determining the image quality. In addition, manpower and costs can be reduced.
[0025]
Hereinafter, a method for mechanically determining the image quality required for each area based on the variance value Vmn for each area divided into, for example, 16 × 16 pixels will be described. A variance value Vmn for determining whether or not an image signal in a region divided into 16 × 16 pixels has a high frequency component is expressed by the following equation (1).
[0026]
[Expression 1]

The variance value Vmn increases according to the amount of high frequency components of the image signal. When the variance value Vmn is smaller than the threshold value Th, it is determined that the area needs to have low image quality, and when the variance value Vmn is larger than the threshold value Th, it is determined that the area requires high image quality. The threshold Th is determined from the relationship with a desired compression rate, quantization table value, and the like. If a plurality of threshold values Th are set and the image quality is determined for each threshold Th, the image quality of the restored image can be finely changed for each region.
[0027]
Here, the predetermined number of pixels in the area to be divided may be constant throughout the image, or the number of pixels to be divided by dividing one area into 32 × 32 pixels and another area into 16 × 16 pixels. May be changed for each region. Also in downsampling described later, downsampling may be performed with an arbitrary number of pixels. Similarly to the case of dividing an area, when JPEG is used as a compression / decompression method, downsampling is performed with an integer multiple of 8 × 8 pixels such as 8 × 8 pixels and 16 × 16 pixels. Processing becomes easy because it is a multiple of 8 × 8 pixels, which is the unit of compression / decompression processing.
[Downsampling section]
For example, when an image is divided into a predetermined number of pixels = 16 × 16 pixels, and 8 × 8 pixels are applied to downsampling a region where low image quality is good,
A region that is divided into 16 × 16 pixels by the dividing unit 110 and a region having a low image quality is down-sampled from 16 × 16 pixels to 8 × 8 pixels by the down-sampling unit 120. On the other hand, the area where high image quality is required remains 16 × 16 pixels and is not down-sampled. Further, the downsampling unit 120 adds the downsampled region and the non-downsampled region and outputs the sum to the encoder 130.
[0028]
Downsampling is performed by thinning processing or filtering processing that extracts pixels in the region. Examples of filter processing include low-pass filter processing and Gaussian filter processing. Downsampling may be performed only by thinning processing. In addition, when downsampling is performed using both filter processing and thinning-out processing, accurate downsampling with less distortion is possible.
For example, when downsampling is performed by Gaussian filter processing and thinning-out processing, processing is performed as follows. The Gaussian filter processing is performed by applying a one-dimensional Gaussian filter in the horizontal direction of the image signal and applying the same filter in the vertical direction to the resulting signal. Further, by performing pixel thinning after the Gaussian filter processing, an image downsampled to a desired size can be obtained.
[0029]
An example of 1/2 downsampling will be described below with reference to FIG. FIG. 2A shows one region of 16 × 16 pixels. As a downsampling method, a downsampled image of 16 × 16 pixel data (hereinafter referred to as a representative value) is stored in an 8 × 8 pixel δs portion (X = 0 to 7, Y = 0 to 7). Fill bits in other αs (X = 8 to 15, Y = 0 to 7), βs (X = 0 to 7, Y = 8 to 15), γs (X = 8 to 15, Y = 8 to 15) regions There is a method of storing “0”. As the fill bit, it is not particularly necessary to use “0”, and another same numerical value can be used. For example, a representative value of 16 × 16 pixel data is stored in a δs portion (X = 0 to 7, Y = 0 to 7) of 8 × 8 pixels, and other αs (X = 8 to 15, Y = 0 to 0). 7), βs (X = 0 to 7, Y = 8 to 15), γs (X = 8 to 15, Y = 8 to 15) The average value, maximum value, and minimum value of 16 × 16 pixels as fill bits Insert the same numerical value.
[0030]
Thus, by downsampling and inserting “0” or the same numerical value into the remaining pixels, the image can be efficiently compressed by entropy coding at the time of coding, and the compression rate can be increased. it can. Further, a flag or the like indicating that it has been downsampled by using the fill bit is unnecessary, and the compression rate is not lowered.
If the number of pixels to be down-sampled is determined for each threshold Th, the image quality of the restored image can be finely changed for each region. For example, in one area, the number of pixels to be downsampled is reduced by downsampling, for example, by downsampling from 32 × 32 pixels to 1/2 of 16 × 16 pixels, and in other areas from 32 × 32 pixels to 8 × 8 pixels You can change it. Also, change the number of pixels to be divided, such as down-sampling one region from 32 × 32 pixels to 1/4 of 8 × 8 pixels and another region from 16 × 16 pixels to 1/2 of 8 × 8 pixels. Downsampling may be performed. In this way, various downsamplings such as 1/2, 1/4, 1/3 and 2/3 can be applied by changing the number of pixels to be divided and the number of pixels to be downsampled or changing them in combination. The image quality of the restored image can be finely changed for each area.
[Encoder, I / F]
The encoder 130 performs compression by a general-purpose compression method using a single quantization table, and the compressed data is transmitted to the decompression side via the I / F 140. An example of a general-purpose compression method using a single quantization table is JPEG.
[0031]
Next, the functional configuration of the decompressing computer 200 will be described.
[Decoder, I / F]
The compressed data is input to the decoder 160 via the I / F 150 and decompressed by a general-purpose decompression method using a single quantization table. The decompression is performed by a method corresponding to the compression method. For example, when the compression is performed by JPEG, the decompression is performed by JPEG.
[Determining part]
The determination unit 170 determines an area that is downsampled. As a method for determining the downsampled region, there is a method for detecting a fill bit.
[0032]
For example, when 16 × 16 pixels are used for division, 8 × 8 pixels are used for downsampling, and “0” is applied to the fill bit, αs, βs, γs when downsampling is performed in FIG. In the area, “0” is stored as fill bits. Therefore, when “0” is stored in all the αs, βs, and γs regions, it is determined that downsampling is performed, and when all are not “0”, it is determined that downsampling is not performed. it can. Whether “0” is stored in all the αs, βs, and γs regions is expressed by the following equation (2) based on FIG.
[0033]
[Expression 2]

Here, when all “0” are stored in the αs, βs, and γs regions, Dmn becomes “0”, and it can be detected that the downsampling is performed.
The number of pixels to be downsampled by applying 32 × 32 pixels for division, 16 × 16 pixels or 8 × 8 pixels corresponding to the image quality of each region for downsampling, and “0” for the fill bit A determination method of downsampling when changing is described with reference to FIGS. 2 (a) and 2 (b). FIG. 2B shows one area of 32 × 32 pixels. First, as shown in FIG. 2B, a region of 32 × 32 pixels is divided into four, and αL (X = 16 to 31, Y = 0 to 15), βL (X = 0 to 15, Y = 16 to 31). , ΓL (X = 16 to 31, Y = 16 to 31), whether or not “0” is stored is determined by the equation (2). Next, when αL, βL, and γL are all “0”, in FIG. 2A, αs (X = 8 to 15, Y = 0 to 7), βs (X = 0 to 7, Y = 8-15) and γs (X = 8-15, Y = 8-15), it is similarly determined whether or not “0” is stored. As a result of such determination, (1) when “0” is not stored in αL, βL, and γL, downsampling is not performed. (2) When “0” is stored in αL, βL, and γL and “0” is not stored in αs, βs, and γs, downsampling is performed from 32 × 32 pixels to 16 × 16 pixels. (3) When “0” is stored in αL, βL, and γL, and “0” is stored in αs, βs, and γs, downsampling is performed from 32 × 32 pixels to 8 × 8 pixels. Can be determined.
[0034]
Furthermore, 32 × 32 pixels or 16 × 16 pixels corresponding to the image quality of each region for division, 8 × 8 pixels for downsampling, and “0” for fill bits are determined for downsampling. The method will be described with reference to FIGS. 2 (a) and 2 (b). First, in FIG. 2A, whether or not “0” is stored in all of αs, βs, and γs out of 16 × 16 pixels is determined by the equation (2). Next, when αs, βs, and γs are all “0”, 16 × 16 pixels including αs, βs, and γs are set as “δL” in FIG. 2B to all of αL, βL, and γL. It is similarly determined whether or not “0” is stored. As a result of such determination, (1) if “0” is not stored in αs, βs, and γs, downsampling is not performed. (2) When “0” is stored in αs, βs, and γs and “0” is not stored in αL, βL, and γL, downsampling is performed from 16 × 16 pixels to 8 × 8 pixels. (3) When “0” is stored in αs, βs, and γs, and “0” is stored in αL, βL, and γL, downsampling is performed from 32 × 32 pixels to 8 × 8 pixels. Can be determined.
[0035]
In addition, various methods for determining downsampling are conceivable. Alternatively, the sampling method may be detected not by detecting that all “0” s are stored in the αs, βs, and γs regions, but by detecting that they are all satisfied with the same numerical value. .
In this way, by detecting that “0” or other same numerical values are stored in all the αs, βs, and γs regions, it is possible to determine the region that is downsampled.
[Interpolation section, output section]
Based on the determination result of the determination unit 170, the interpolation unit 180 interpolates the downsampled region to the original number of pixels, such as 8 × 8 pixels to 16 × 16 pixels. On the other hand, interpolation is not required for regions that are not downsampled. In addition, the interpolation unit 180 adds the region interpolated after downsampling and the region not downsampled and not interpolated, and outputs the sum to the output unit 190. The output unit 190 converts the image data so that it can be displayed in accordance with the number of display pixels of an LCD (not shown), for example. Therefore, it is possible to interpolate the pixels in the downsampled area based on the data after downsampling and output the image of the area. Further, since there is no need to interpolate an image in a region that has not been downsampled, it is possible to output a higher quality image than an image in a region that has been downsampled.
[0036]
As an interpolation method, for example, an LP enlargement method capable of estimating a high-frequency component of a lost image, linear interpolation, quadratic interpolation, cubic interpolation, or the like can be used. The LP enlargement method is a method for reconstructing an image by estimating a high-frequency component lost during downsampling by the downsampling unit 120.
Since the compression / decompression system according to the present invention performs downsampling according to the image quality required for each area before encoding an image composed of a plurality of areas, the image compression / decompression is performed using a single quantization table. The image quality can be changed for each area. Therefore, using a general-purpose format using a single quantization table, image compression / decompression can be performed while changing the image quality for each region. Also, by downsampling at least one area according to the image quality of each area, it is possible to increase the compression efficiency of the entire image compared to the case of simply compressing, and in the area where high image quality is required Can be output. If the number of pixels to be down-sampled is changed for each region, the degree of freedom in changing the image quality can be increased. By increasing the compression efficiency, the transmission efficiency by the network 300 can be increased.
[0037]
Further, the dividing unit 110 and the downsampling unit 120 can be taken into the encoder 130. Similarly, the determination unit 170 and the interpolation unit 180 can be taken into the decoder 160. In this way, by incorporating various processes into the encoder 130 and the decoder 160, the configuration is simplified, and the processing speed can be improved, the device can be downsized, and the cost can be reduced.
Further, in the compression / decompression system shown in FIG. 1, the compression-side computer 100 and the decompression-side computer 200 are connected via a network 300, but compression / decompression is performed by the same computer and is not performed via the network 300. -It may be an expansion system. As described above, even when the network is not used, the capacity for storing images can be reduced, and the number of stored images can be increased.
<Performance evaluation>
Next, evaluation of an image compressed / expanded by the image compression / expansion system according to the embodiment will be described. In the evaluation, JPEG is used as a compression / decompression method.
[0038]
Examples of a method for determining the quality of image quality by image compression / decompression include the following methods. One of them is a method of numerical evaluation using a graph in which the horizontal axis represents the file size (compression ratio) and the vertical axis represents PSNR (Peak Signal to Noise Ratio). This makes it possible to evaluate the quality of an image with the same file size, and conversely to evaluate the file size with the same image quality by numerical values. Here, PSNR represents a measure of image degradation due to encoding. That is, it is a scale for evaluating the difference between the image before compression and the restored image after compression / decompression, and the image deteriorates as the PSNR decreases. The other method is a method that evaluates subjectively rather than evaluating image quality by numerical evaluation.
[0039]
Below, the result of having evaluated image quality by the evaluation method using PSNR and the result of having evaluated by subjectivity are demonstrated. The original image before compression / decompression used is an 8-bit gray scale image having a size of 256 × 256 (see the original image in FIG. 6).
First, the result by the evaluation method using PSNR will be described. The contents of the three types of experiments evaluated in Table 1 are shown, and the specific contents of the three types of experiments in Table 1 are shown below.
[0040]
[Table 1]

(1) Downsampling 1 + JPEG: When the original image is divided into 16 × 16 pixel regions, and the variance value Vmn of the region is smaller than the threshold Th (Vmn <Th), downsampling to 8 × 8 pixels and the variance value Vmn Is greater than the threshold Th (Vmn> Th), downsampling is not performed. The image data after the downsampling is JPEG encoded. FIG. 3 shows a state in which an image is divided into 16 × 16 pixels, and a fill bit is inserted into a region that is downsampled to 8 × 8 pixels among the divided regions (region). A). Further, no fill bit is inserted in a region that is not down-sampled (see region B).
(2) Downsampling 2 + JPEG: When the original image is divided into 32 × 32 pixel regions, and the variance value Vmn of the region is smaller than the threshold Th2 (Vmn <Th2), downsampling to 8 × 8 pixels and the variance value Vmn Is larger than the threshold Th2 and smaller than the threshold Th1 (Th2 <Vmn <Th1), downsampling to 16 × 16 pixels, and when the variance value Vmn is larger than the threshold Th1 (Vmn> Th1), no downsampling is performed. . The image data after the downsampling is JPEG encoded. FIG. 4 shows a state in which an image is divided into 32 × 32 pixels, and fill bits are inserted into regions that are downsampled to 8 × 8 pixels or 16 × 16 pixels among the divided regions. (See region C or D). Further, no fill bit is inserted in a region not down-sampled (see region E).
(3) JPEG: JPEG encoding is performed without downsampling the original image.
[0041]
The experiment (3) is an experiment for comparison with the experiments (1) and (2) to which the present invention is applied.
FIG. 5 shows the results of evaluating the experimental results in Table 1 by PSNR with respect to the bit rate (compression rate). As shown in FIG. 5, in all cases where the bit rate is 0.2 to 0.5 (bit / pixel), the PSNR of (1) and (2) to which the present invention is applied does not apply the present invention (3 ) Greater than PSNR. Therefore, in the evaluation of the image quality by PSNR, it can be evaluated that the image compression / decompression is excellent when the present invention is applied. In particular, the method of changing the number of pixels for downsampling for each threshold ((2) downsampling 2 + JPEG) has a high PSNR, that is, an excellent image because there is little difference between the image before compression and the restored image after compression / decompression. It can be evaluated that it has quality.
[0042]
Next, in order to perform subjective evaluation, FIGS. 6 to 9 show images after image compression / decompression by the above-described experiment. 6 to 9 are restored images when compression rates of 0.5, 0.4, 0.3, and 0.2 (bit / pixel) are applied in order.
In all of FIGS. 6 to 9, the downsampling for each region of the present invention is applied to the regions (1) and (2) where only the JPEG encoding is applied (3) (FIG. 3). 4 that the image of the region B or the region E) is clear. In addition, when the number of pixels for downsampling is changed for each threshold (2), the image quality of a region that is not downsampled is better than when (1) is downsampled with a fixed number of pixels. It can also be seen that the higher the image compression rate, the better the image quality of the area (area B or area E) where high image quality is required in the order of (3) <(1) <(2).
[0043]
Next, FIG. 10 and FIG. 11 show experimental results when the region to be downsampled and the region not to be sampled are manually divided to determine the number of pixels for downsampling. FIG. 10 shows a case where the human part and the background part are manually divided, and then the background part is downsampled from 16 × 16 pixels to 8 × 8 pixels, respectively, and the compression rate is set to 0.4 (bit / pixel). And an image when the compression ratio is set to 0.4 (bit / pixel) only by JPEG encoding. On the other hand, FIG. 11 shows that when the background portion is down-sampled from 32 × 32 pixels to 8 × 8 pixels and the compression rate is 0.3 (bit / pixel), the compression rate is 0 only by JPEG encoding. .3 (bit / pixel) image. In both FIG. 10 and FIG. 11, it can be seen that the image of the portion of the person that is not down-sampled is clearer when the down-sampling according to the present invention is applied than with the JPEG-only encoding.
[0044]
From the above results, if the downsampling for each area of the present invention is performed mechanically or manually, the image quality of the required area can be improved over the image quality of the other areas even at the same compression rate. That is, a high compression rate can be obtained with the same image quality.
[Application of the present invention to others]
The present invention can also be used to change the image quality for areas with changes in motion. For example, by increasing the image quality only in an area where there is a change in motion and lowering the image quality in other areas, the overall compression rate can be increased without reducing the image quality in the area where there is a change in motion.
[0045]
Furthermore, in the present invention, the image quality may be changed for each individual image area with respect to an image in which a plurality of images are combined and displayed on one screen. For example, images of a plurality of points are acquired by a plurality of monitoring cameras, and a plurality of images for each camera are arranged as described above to form one image and displayed on the monitoring monitor. When compressing this image, it is possible to change the image quality of the image for each camera by receiving the image quality setting for the image area for each camera. At this time, the image quality of the camera image in which the motion has changed may be increased and the image quality of other images may be automatically decreased by a sensor in the vicinity of the camera.
[0046]
【The invention's effect】
By using the present invention, it is possible to change the image quality for each area while using a general-purpose format using a single quantization table.
[Brief description of the drawings]
FIG. 1 shows an example of an image compression / decompression system according to an embodiment of the present invention.
FIG. 2A is a region of 16 × 16 pixels.
(B) One area of 32 × 32 pixels.
FIG. 3 shows a state in which an image is divided into 16 × 16 pixels and fill bits are inserted.
FIG. 4 shows an image divided into 32 × 32 pixels and a fill bit inserted.
FIG. 5 shows the result of evaluating the experimental results in Table 1 by PSNR with respect to the bit rate.
FIG. 6 shows a restored image when a compression ratio of 0.5 (bit / pixel) is applied.
FIG. 7 shows a restored image when a compression rate of 0.4 (bit / pixel) is applied.
FIG. 8 shows a restored image when a compression rate of 0.3 (bit / pixel) is applied.
FIG. 9 shows a restored image when a compression rate of 0.2 (bit / pixel) is applied.
FIG. 10 shows an image when the background portion is down-sampled and the compression ratio is 0.4 (bit / pixel) among the human portion and the background portion.
FIG. 11 shows an image when the background portion is down-sampled and the compression ratio is set to 0.3 (bit / pixel) among the human portion and the background portion.
FIG. 12 shows an example of a compression / decompression system.
FIG. 13 is a quantization table for each luminance (Y) and color difference (UV).
FIG. 14 shows an example of a compression / decompression system when image quality is changed for each region using a quantization table that is different for each region.
[Explanation of symbols]
1: Compressor
2: Stretcher
10, 105: Input unit
20: DCT converter
30: Quantizer
35, 65: quantization table
40: Encoder
50: Decoder
60: Inverse quantizer
70: Inverse DCT converter
80, 190: Output unit
100, 200: Computer
110: Dividing part
120: Downsampling unit
130: Encoder
140, 150: I / F
160: Decoder
170: determination unit
180: Interpolation unit
300: Network

Claims (12)

A division step of dividing the image into a plurality of regions;
At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least a part of the area. A downsampling step to insert fill bits in the rest of the region;
A quantization step of quantizing the image that has undergone the division step and the downsampling step with a common quantization table for the plurality of regions;
An encoding step for encoding the image quantized in the quantization step;
A decoding step of decoding the image encoded in the encoding step;
An inverse quantization step of inversely quantizing the image decoded in the decoding step with the quantization table;
An interpolation step of interpolating pixels in a region that is down-sampled among a plurality of regions constituting the image that has been inversely quantized in the inverse quantization step;
An image output step of outputting the image with an image quality required for each of the plurality of regions;
Image compression / decompression method. A division step of dividing the image into a plurality of regions;
At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least a part of the area. A downsampling step to insert fill bits in the rest of the region;
A quantization step of quantizing the image that has undergone the division step and the downsampling step with a common quantization table for the plurality of regions;
An encoding step for encoding the image quantized in the quantization step;
An image compression method including: The image compression method according to claim 2, wherein the quantization step and the encoding step are performed by JPEG (Joint Photographic Expert Group). The dividing step includes
A division instruction step for receiving an instruction to divide the image into the plurality of regions;
An image quality receiving step of receiving an image quality setting required for at least one of the plurality of areas;
The image compression method according to claim 2, comprising: The dividing step includes
An image dividing step of dividing the image into the plurality of regions including a predetermined number of pixels;
An image quality determination step for determining an image quality required for each region according to a signal amount indicating a high-frequency component for each of the plurality of regions;
The image compression method according to claim 2, comprising: A dividing means for dividing the image into a plurality of regions;
At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least a part of the area. Downsampling means for inserting fill bits in other parts of the region;
A quantizer that quantizes the image that has passed through the dividing unit and the downsampling unit using a common quantization table for the plurality of regions;
An encoder that encodes an image quantized in the quantizer;
An image compression apparatus including: A dividing means for dividing the image into a plurality of regions;
At least one of a plurality of areas constituting the image is down-sampled according to the image quality required for the area, and the down-sampled image (hereinafter, representative value) of the area is at least a part of the area. Down-sampling means to insert fill bits in other parts of the region,
Quantizing means for quantizing the image that has passed through the dividing means and the downsampling means with a common quantization table for the plurality of regions; and
Encoding means for encoding the image quantized in the quantization means;
Image compression program to make the computer function as A decoding step of decoding an image that is divided into a plurality of regions, and is quantized by a common quantization table for the plurality of regions after at least one of the plurality of regions is downsampled. When,
An inverse quantization step of inversely quantizing the image decoded in the decoding step with the quantization table;
An interpolation step of interpolating pixels in a region that is down-sampled among a plurality of regions constituting the image that has been inversely quantized in the inverse quantization step;
An image output step of outputting the image with an image quality required for each of the plurality of regions;
An image decompression method including: The image decompression method according to claim 8, wherein the decoding step and the inverse quantization step are performed by JPEG. 9. The image decompression method according to claim 8, wherein the interpolation step further includes a step of specifying a region that is down-sampled based on whether or not a predetermined fill bit is inserted. A decoder that decodes an image that is divided into a plurality of regions, and is quantized by a common quantization table for the plurality of regions after at least one of the plurality of regions is downsampled. When,
An inverse quantizer for inversely quantizing an image decoded by the decoder using the quantization table;
Interpolating means for interpolating pixels in a region that is down-sampled among a plurality of regions constituting an image quantized inversely in the inverse quantizer;
Image output means for outputting the image with image quality required for each of the plurality of regions;
An image expansion device including: Decoding means for dividing an image that is divided into a plurality of regions and is quantized by a common quantization table for the plurality of regions after at least one of the plurality of regions is downsampled ,
Inverse quantization means for inversely quantizing the image decoded by the decoding means using the quantization table;
Interpolating means for interpolating the pixels of the downsampled region among the plurality of regions constituting the image inversely quantized by the dequantizing means; and
An image expansion program for causing a computer to function as image output means for outputting the image with image quality required for each of the plurality of areas.

Images