JP3572205B2 - How to identify circular patterns - Google Patents

Description

【００２１】
求められる値、つまり２つのベクトルの内積を各ベクトルの大きさの積で除したものである。このように求められる相関係数ｒは両ベクトルの類似度を表しており、まったく異なることを示す０から同一を示す１までの間の値を示すものである。なお、ｒ_０はＦ_０とＦ_０との相関係数であるから、同一を示す１となる。
【００２２】
相関係数ｒ_１を求めたら、更に１画素シフトして画像データＤ_２を表すベクトルＦ_２＝（ｆ_１２６，ｆ_１２７，ｆ_０，ｆ_１，…，ｆ_１２５）を求め、Ｆ_０とＦ_２との相関係数ｒ_２を求める。そして、このようなシフトと演算を続けて行い、ｒ_６４までを求める。
【００２３】
このようにして求められたｒ_０ 〜ｒ_６４の相関係数ｒは、図３に示すように、グラフに相関データ２として表される。ここで、画像データＤを１２８回シフトすれば１２８個の相関係数ｒ_０ 〜ｒ_１２７が出力されることになるが、ｒ_６５〜ｒ_１２７の相関係数は、グラフに表したとしても、ｒ_６４を境としてｒ_０ 〜ｒ_６３と対称なグラフになるため、このｒ_６５から先はデータをとる必要がない。したがって、上述した相関データ２は、円形パターン１の画像データＤの１画素ずつのシフトを半分まで行って構成することとしている。
【００２４】
そして、図３に示される相関データ２を一つのベクトルＳ＝（ｒ_０，ｒ_１，ｒ_２，…，ｒ_６４）で表すとすると、このベクトルＳは、外形が円形である識別対称物の種類ごとに固有の特徴を有するベクトルとなるから、いいかえれば種類ごとに固有の傾向を示す波形が得られることとなる。
【００２５】
また、このようにして得られる配列データは、回転にまったく影響を受けない特徴量である。これは、数式１で示したように、相関係数ｒを求めるときの分子の内積、分母の大きさの積とも回転による影響を受けないものであり、この２つの比である相関係数ｒももちろん対象物の回転位置とは無関係であることによる。したがって、円形パターン１のどの位置から画像データＤを切り出したとしてもまったく同一の相関データ２を得ることができる。
【００２６】
相関係数ｒを求める式の分子、分母ともに回転による影響を受けないことについて詳述すれば、以下のようになる。例えば数式１における分子のＦ_０ ・Ｆ_１ は、Ｆ_０とＦ_１の内積を表す量であり、同式のように、この内積はそれぞれ対応する成分の積の和という形で求められている。そして、仮に、対象物が回転してしまっていて、異なる位置から画像データＤが切り出されていたとした場合、Ｆ_０、Ｆ_１とも同量ずつ画素がずれることとなる。例えばＦ_０が３画素ずれていてＦ_０′＝（ｆ_３，ｆ_４，ｆ_５，ｆ_６，…，ｆ_２）だとすれば、Ｆ_１′＝（ｆ_２，ｆ_３，ｆ_４，ｆ_５，…，ｆ_１）となる。そして両者の内積は、数式２のようになる。
【００２７】
【数２】

この場合、画素順はずれているが、Ｆ_０とＦ_１の各成分の対応関係には変化がない。したがって、この結果として得られるｒ_１に影響はなく、まったく同一の変化量を得ることができる。
【００２８】
以上が、１つの識別対象物から円形パターン１を利用して相関データ２を得るまでの過程である。
【００２９】
なお、上述した方法では、１２８×１画素のラインを１本とった場合について説明したが、複数のラインを抽出することもできる。即ち、例えば図４に示すように、１２８×１０画素の２次元配列の画像データＤを得た場合にあっては、この画像データＤを１２８×１のラインデータ３，…，３が１０本まとまったものとみなし、上述したように、各ラインデータ３，…，３について１画素ずつシフトさせながらもとのラインデータ３，…，３との相関係数ｒをその都度求めていくようにすれば良い。この場合、相関係数ｒとシフト量との関係を示す相関データ２，…，２が、図５に示すように全部で１０本得られる。
【００３０】
次に、種類判別の基準となるテンプレートを作成し、この作成されたテンプレートを利用して円形パターン１を識別する場合について説明する。
【００３１】
円形パターン１の画像データＤを１画素ずつシフトさせてもとの画像データＤ_０ との相関係数ｒを求めるまでは上述した過程と同じであり、このように相関係数ｒを求めて複数の相関データ２，…，２が得られたら、これらをサンプルとして用意しておく（図６）。そして、用意されたサンプルデータをもとに、その分類カテゴリーの特徴を的確に表す代表的な配列データを決定し、基準のテンプレート４として用意しておく。
【００３２】
識別対称となる対象物つまり円形プレート１がどの種類に分類されるかは、この基準テンプレート４を用いて行う。つまり、基準テンプレート４は各種類別に作成された代表的な配列データであり、この配列との類似度（あるいは非類似度）を求めて円形パターン１の種類を分類することができる。
【００３３】
なお、このようにして代表的な配列データを決定するにあたっては、得られた配列データを１次元配列のベクトルとみなし、主成分分析を用いて基準テンプレート４となる代表的なベクトルを決定するようにしても良い。
【００３４】
ここまで説明した本発明の識別方法を用いれば、以下のように、少ない処理時間で円形プレート１を識別できるようになる。即ち、これまで実用されている識別方法では、画像データＤが１２８×１０の画素からなれば、その画像全体を横方向に１２８回シフトさせなければならなかったのに対し、本発明によれば、相関係数ｒを求める計算を（６４＋１）回、つまりベクトルＳの演算回数６４回と、基準テンプレート４とのマッチング回数１回で済ますことができる。マッチング回数１回はテンプレートの枚数だけ必要となるものであるが、ベクトルＳの演算回数に比して少ないものである。
【００３５】
なお、上述の実施形態は本発明の好適な実施の一例ではあるがこれに限定されるものではなく本発明の要旨を逸脱しない範囲において種々変形実施可能である。例えば本実施形態で説明した識別方法において、エッジ強調フィルタを使用することによって、グラフの増減を強調してより明確に識別できるようにしても良い。例えば図７に示すように、ｒ_２を基準とした前後２つずつの画素ｒ_０，ｒ_１，ｒ_３，ｒ_４に対してそれぞれ１，１，−１，−１を乗じ、和をとってｒ′_２ とする。このとき、ｒ_２より左側（ｒ_１のある側）が小さく、右側（ｒ_３のある側）が急激に大きくなっていればｒ′_２はマイナス方向に大きく振れることとなる。一方、ｒ_２より左側が大きく、右側が急激に小さくなっていればｒ′_２は逆にプラス方向へ大きく振れることとなる。また、ｒ_２を中心に左右の値に差がない場合、ｒ′_２は０に近い値となる。この結果、左右の値が大きく変化するエッジ部分が強調され、相関係数ｒの変化の様子が認識し易くなる。
【００３６】
また、本発明の応用としては、例えば、経験的に同一種類のサンプルのベクトルを複数重ね合わせて平均をとり、その代表波形をテンプレートとしても良いし、ニューロ技術の一手法であるＬＶＱを用いて代表波形を求め、これをテンプレートとしても良い。
【００３７】
また、基準テンプレート４となるべきベクトルが、より代表的なベクトルに類似したものになるように、この代表的なベクトルを基準としてレベル変換を行うようにしても良い。この変換により排除すべきベクトルはより類似度が低くなり、排除が容易になる。
【００３８】
【数３】

【００３９】
また、相関係数ｒを求める際には、数式３を用いて、正確で変動が少ない計算を行うようにしても良い。ここで、この式を用いたときの相関係数をｒ_ｇ、これ以外の式を用いたときの相関係数ｒ_ｂとすると、この２つのうちのどちらを適用するかについては図８に示すように４種類の組み合わせが考えられ、◎部分が有効的である。なお、ｒ_ｂ−ｒ_ｂ（ともにｒ_ｂ）の場合は計算式に変動要素が入るためマッチング結果のばらつきが大きくなり、これに対しては上述したレベル変換を行うことにより変動要素をなくすことができる。また、ｒ_ｇ−ｒ_ｇの場合は結果に特に問題ないが、原則的にはｒ_ｇ−ｒ_ｂ、ｒ_ｂ−ｒ_ｇ、同じ結果になると考えられるので、計算量が多くなる点で他の組み合わせに比べ不利になる。
【００４０】
また、類似度を計算する方法として、例えばベクトル間のユークリッド距離ｄを求めて計算する方法があり、本実施形態において説明したような相関係数を求める方法に特に限定されることはない。
【００４１】
また、隣接しているか否かにかかわらず、複数のリングを切り出して上述した実施形態と同様の処理を行うようにしても良い。これによれば、重要な特徴が含まれていると思われる部分を抽出することも可能である。
【００４２】
【発明の効果】
以上の説明より明らかなように、請求項１記載の発明によれば、円形パターンを円周上のどの位置から切り出したとしても相関係数は変化しないから、円形パターンの位置合わせ作業が不要となり、円形対象物の回転量に影響されることなく相関データを求めることができる。したがって、円形パターンを識別するときに要する処理時間を短縮することができる。
【００４３】
また請求項２記載の発明では、画像データを１画素ずつシフトして半回転すれば１回転分の相関データを求められるから、対象物を識別するときの処理すべき情報量が少なく、処理時間が短縮できる。
【００４４】
更に請求項３記載の発明によれば、円形パターンを複数のラインデータから構成しているので、多くの相関データを用いて対象物を識別することが可能であり、しかも、処理すべき情報量を少なくすることができる。
【００４５】
請求項４記載の発明によれば、円形パターンの位置合わせが不要なことに加え、画像データを半回転シフトすれば相関データを得られることから、テンプレートの作成が容易である。またこれにより、基準のテンプレートを用いて円形パターンを識別するときの処理時間を短縮できる。
【００４６】
また請求項５記載の発明では、円形パターンの識別に必要となるテンプレートの用意作業が容易であり、これらを用いて対象物の円形パターンを正確に識別することができる。
【００４７】
そして請求項６記載の発明では、分類カテゴリーにおけるパターンを特徴的に表すものを基準テンプレートとして採用しているから、対象物である円形パターンを正確に識別することができる。
【図面の簡単な説明】
【図１】本発明の一実施形態を示す図であり、円形パターンから画像の一部をリング状に切り出して画像データとする様子を示す。
【図２】画像データを１画素ずつシフトさせ、それぞれについてもとのデータとの相関係数ｒを求める様子を示す図である。
【図３】求めた相関係数とシフト量との関係を示す相関データの一例を示す図である。
【図４】円形パターンから画像の一部をリング状に切り出し、画像データとして複数のラインを抽出する様子を示す図である。
【図５】１０ラインからなる相関データの一例を示す図である。
【図６】複数のサンプルデータの中から基準のテンプレートを選択する様子を示す図である。
【図７】エッジ強調フィルタによるエッジ強調の様子を示す図である。
【図８】相関係数ｒ_ｂとｒ_ｇとの組み合わせの有効性を示す表である。
【符号の説明】
１ 円形パターン
２ 相関データ
３ ラインデータ
４ 基準のテンプレート
Ｄ 画像データ
Ｄ_０ もとの画像データ
ｒ 相関係数[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for identifying a circular pattern. More specifically, the present invention relates to an improvement in a method for identifying a circular pattern used to identify an object having a particularly circular outer shape.
[0002]
[Prior art]
In order to identify an object having a circular outer shape, the following identification method using an image of a circular pattern has been put to practical use.
[0003]
First, a part of the image is extracted in the form of a circle or a ring from the center of the circular pattern of the object, developed into a band image (two-dimensional array), and necessary processing is performed on the data to convert data of a certain size. obtain. Then, the obtained fixed-size image is subjected to pattern matching with a reference template image of the same size, and a similarity to this template is obtained to identify a target.
[0004]
When a circular object is to be identified as described above, the template and the image to be identified do not always have the same inclination because of the rotation of the image due to rotation. Then, the tilt that occurs without being fully adjusted appears as a left-right shift in the extracted band-shaped image (two-dimensional array). Therefore, in such a case, pattern matching is performed on the template each time while shifting the determination target image one pixel at a time in the horizontal direction, and a shift position having the highest similarity is selected from the matching results shifted from 0 to 2π. We find out and identify based on the result.
[0005]
[Problems to be solved by the invention]
However, in the method of performing pattern matching while shifting each target image as described above, the entire image must be shifted to 2π by one pixel with respect to the angular coordinates. There is a problem that the target processing speed cannot be cleared because the time increases.
[0006]
Accordingly, an object of the present invention is to provide a method for identifying a circular pattern that can reduce the amount of information to be processed and shorten the processing time. Another object of the present invention is to provide a method for identifying a circular pattern using a template that is easy to create.
[0007]
[Means for Solving the Problems]
In order to achieve this object, according to the first aspect of the present invention, image data of a circular pattern is read, and the read image data is shifted one pixel at a time to obtain a correlation coefficient with the original data. The circular pattern is identified by using the correlation data constituted by the relation numbers as the identification data of the circular pattern.
[0008]
Here, the image data is represented as a vector having components by the number of pixels, and the correlation coefficient is represented by a ratio of an inner product of two image data and a product of the size of each image data. In such a case, even if the circular pattern is cut from any position on the circumference, the correlation coefficient does not change, so that it is not necessary to align the circular pattern, which affects the rotation amount of the circular object. The correlation data can be obtained without being performed.
[0009]
According to a second aspect of the present invention, in the method for identifying a circular pattern according to the first aspect, the correlation data is configured by shifting the image data of the circular pattern by one pixel at a time by half. In this case, the correlation data obtained from the correlation coefficient becomes symmetrical data after the image data is shifted one pixel at a time and rotated by half a rotation. Can be requested. Therefore, the amount of information to be processed when identifying an object is small.
[0010]
According to a third aspect of the present invention, in the method for identifying a circular pattern according to the first or second aspect, the circular pattern is data of a two-dimensional array composed of a plurality of sets of line data. For comparison data. Therefore, in this case, it is possible to identify the object using a lot of correlation data, and nevertheless, the amount of information to be processed is small.
[0011]
In the method for identifying a circular pattern according to the fourth aspect of the present invention, the image data of the circular pattern is shifted one pixel at a time to obtain a correlation coefficient with the original data, and the correlation data constituted by the correlation coefficient is used as a reference. The image data of the circular pattern to be compared is shifted by one pixel to obtain a correlation coefficient with the original data of the pattern to be compared, and the correlation data constituted by the correlation coefficient is prepared. Is compared with the template to identify the circular pattern.
[0012]
In this case, even if the circular pattern is cut from any position on the circumference, the correlation coefficient does not change, so that it is not necessary to perform the alignment, and the correlation data obtained from the correlation coefficient is obtained by dividing the image data one pixel at a time. Since it can be obtained by shifting by half a rotation, a reference template can be easily created.
[0013]
According to a fifth aspect of the present invention, in the method for identifying a circular pattern according to the fourth aspect, a reference template is prepared for each type of circular pattern to be classified. In this case, since the creation of the reference template is easy, the work of preparing the number of templates necessary for identifying the circular pattern is also simplified.
[0014]
According to a sixth aspect of the present invention, in the method for identifying a circular pattern according to the fourth or fifth aspect, a plurality of correlation data are obtained as samples, and a representative one which accurately indicates the characteristics of the circular pattern is obtained by using them. , Which is used as a reference template. As described above, since the reference template is selected as a characteristic of the pattern in the classification category, a circular pattern as an object can be accurately identified by using the reference template.
[0015]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, the configuration of the present invention will be described in detail based on an example of an embodiment shown in the drawings.
[0016]
1 to 6 show an embodiment of a method for identifying a circular pattern according to the present invention. The present invention reads image data D of a circular pattern 1 as an object, shifts the read image data D one pixel at a time to obtain a correlation coefficient r with the original image data D ₀ , The circular pattern 1 is identified by using the correlation data 2 constituted by the relation number r as the identification data of the circular pattern 1.
[0017]
First, as shown in FIG. 1 or FIG. 4, the image data of the circular pattern 1 is read, the center of the circular pattern 1 is detected, and a part of the image is formed in a circular or ring shape at an equal distance from the center of the circular pattern 1. The image data D is extracted and developed into a band image (one-dimensional or two-dimensional array) to form image data D. If the number of rings to be extracted is one, it is developed into linear data (in this case, a one-dimensional array), and necessary processing is performed on this data to obtain a certain size. In this embodiment, in FIG. 1, one line composed of 128 pixels in a horizontal row is cut out in a ring shape and developed as band-shaped data, and in FIG. 4, a plurality of lines are cut out in a ring shape and developed as two-dimensional band-shaped data. I have.
[0018]
And regarded as thus clipped image data D to 128-dimensional vector composed of 128 pixels, it is shifted by one pixel while obtaining a correlation coefficient r of the original image data D _0. That is, as shown in FIG. 2, the original image data D _0, that is, the image data D obtained by shifting one pixel from the vector F ₀ = (f ₀ , f ₁ , f ₂ , f ₃ ,..., F ₁₂₇ ) ₁ (in this case, the pixel f ₁₂₇ at the right end is brought to the head at the left end), and a vector F ₁ = (f ₁₂₇ , f ₀ , f ₁ , f ₂ ,..., F ₁₂₆ ) is obtained, and F ₀ and obtaining a correlation coefficient _{r 1} of the F _1.
[0019]
This correlation coefficient r ₁ is expressed by the following equation (1).
(Equation 1)

[0021]
The value to be obtained, that is, the inner product of two vectors divided by the product of the magnitudes of the vectors. The correlation coefficient r thus obtained indicates the similarity between the two vectors, and indicates a value between 0 indicating completely different and 1 indicating the same. Since r ₀ is a correlation coefficient between F ₀ and F ₀ , it is 1 indicating the same.
[0022]
Once the correlation coefficient r ₁ is obtained, the vector F ₂ = (f ₁₂₆ , f ₁₂₇ , f ₀ , f ₁ ,..., F ₁₂₅ ) representing the image data D ₂ is further shifted by one pixel, and F ₀ and F 2 are obtained. obtaining a correlation coefficient _{r 2} and _2. Then, it performs continued operation with such a shift, seeking to r _64.
[0023]
The correlation coefficients r of r ₀ to r ₆₄ thus obtained are represented as correlation data 2 in a graph as shown in FIG. Here, if the image data D is shifted 128 times, 128 correlation coefficients r _{0 to} r ₁₂₇ will be output. Even if the correlation coefficients of r _{65 to} r ₁₂₇ are represented in a graph, Since the graph is symmetrical with r ₀ to r ₆₃ with r ₆₄ as a boundary, there is no need to take data after r ₆₅ . Therefore, the above-described correlation data 2 is configured by shifting the image data D of the circular pattern 1 by one pixel at a time by half.
[0024]
If the correlation data 2 shown in FIG. 3 is expressed by one vector S = (r ₀ , r ₁ , r ₂ ,..., R ₆₄ ), this vector S is Since the vector has a characteristic unique to each type, in other words, a waveform indicating a unique tendency for each type can be obtained.
[0025]
The array data obtained in this way is a feature amount that is not affected by rotation at all. This is because, as shown in Expression 1, neither the inner product of the numerator nor the product of the size of the denominator when calculating the correlation coefficient r is affected by rotation, and the correlation coefficient r, which is the ratio of the two, is obtained. This is, of course, independent of the rotational position of the object. Therefore, even if the image data D is cut out from any position of the circular pattern 1, exactly the same correlation data 2 can be obtained.
[0026]
The fact that neither the numerator nor the denominator of the equation for calculating the correlation coefficient r is affected by rotation will be described in detail below. For example F ₀ · F ₁ molecules in Equation 1 is an amount representing the inner product of F ₀ and F _1, as shown in the equation, the inner product is obtained in the form of sum of products of the corresponding components . If the object is rotated and the image data D is cut out from a different position, the pixels of F ₀ and F ₁ are shifted by the same amount. For example, if F ₀ is shifted by 3 pixels and F ₀ ′ = (f ₃ , f ₄ , f ₅ , f ₆ ,..., F ₂ ), then F ₁ ′ = (f ₂ , f ₃ , f ₄ , f ₅ ,..., f ₁ ). The inner product of the two is as shown in Expression 2.
[0027]
(Equation 2)

In this case, although the pixel order is shifted, there is no change in the correspondence between the components F ₀ and F ₁ . Therefore, there is no effect on r ₁ obtained as a result, and exactly the same amount of change can be obtained.
[0028]
The above is the process up to obtaining the correlation data 2 from one identification target using the circular pattern 1.
[0029]
Note that, in the above-described method, a case where one line of 128 × 1 pixels is used has been described, but a plurality of lines can be extracted. That is, as shown in FIG. 4, for example, when image data D in a two-dimensional array of 128 × 10 pixels is obtained, this image data D is divided into ten 128 × 1 line data 3,. Assuming that the data is a unit, as described above, the correlation coefficient r with the original line data 3,..., 3 is obtained each time while shifting each line data 3,. Just do it. In this case, a total of ten pieces of correlation data 2,..., 2 indicating the relationship between the correlation coefficient r and the shift amount are obtained as shown in FIG.
[0030]
Next, a case will be described in which a template serving as a reference for the type determination is created, and the circular pattern 1 is identified using the created template.
[0031]
The process until the correlation coefficient r with the original image data D ₀ is obtained by shifting the image data D of the circular pattern 1 by one pixel is the same as the above-described process. Are obtained as samples (FIG. 6). Then, based on the prepared sample data, representative sequence data that accurately represents the characteristics of the classification category is determined and prepared as a reference template 4.
[0032]
The type of the object to be identified, that is, the circular plate 1 is classified using the reference template 4. That is, the reference template 4 is representative array data created for each type, and the type of the circular pattern 1 can be classified by calculating the similarity (or dissimilarity) with this array.
[0033]
In determining representative array data in this manner, the obtained array data is regarded as a one-dimensional array vector, and a representative vector serving as a reference template 4 is determined using principal component analysis. You may do it.
[0034]
By using the identification method of the present invention described above, the circular plate 1 can be identified in a short processing time as described below. That is, according to the identification method that has been practically used, if the image data D is composed of 128 × 10 pixels, the entire image must be shifted 128 times in the horizontal direction. , The calculation of the correlation coefficient r can be performed by (64 + 1) times, that is, 64 times of the operation of the vector S and one time of the matching with the reference template 4. Although the number of times of matching is required by the number of templates, the number of times of matching is smaller than the number of times of calculation of the vector S.
[0035]
The above embodiment is an example of a preferred embodiment of the present invention, but the present invention is not limited to this, and various modifications can be made without departing from the gist of the present invention. For example, in the identification method described in the present embodiment, an edge enhancement filter may be used to emphasize the increase or decrease of the graph so that the graph can be more clearly identified. For example, as shown in FIG. 7, a pixel _r 0 of the two by two front and rear relative to the _{r 2,} r _1, r 3, respectively _{r 4} 1, 1, -1, multiplied by -1, summed R ′ ₂ . At this time, (a certain side r ₁₎ is less than r ₂ left, r _'2 If is sufficient that rapidly increased (by the r ₃₎ right so that the largely fluctuates in the negative direction. On the other hand, larger left than r _2, r _'2 if right sharply smaller so that the largely fluctuates in the opposite to the positive direction. If there is no difference between the left and right values around the r _2, r _'2 is a value close to 0. As a result, the edge portion where the left and right values greatly change is emphasized, and the state of the change of the correlation coefficient r is easily recognized.
[0036]
In addition, as an application of the present invention, for example, a plurality of vectors of the same type of sample are empirically superimposed and averaged, a representative waveform thereof may be used as a template, or LVQ which is a method of neurotechnology may be used. A representative waveform may be obtained and used as a template.
[0037]
Further, the level conversion may be performed using the representative vector as a reference so that the vector to be the reference template 4 becomes more similar to the representative vector. Vectors to be eliminated by this conversion have lower similarity, and are easily eliminated.
[0038]
(Equation 3)

[0039]
Further, when calculating the correlation coefficient r, an accurate calculation with little change may be performed by using Expression 3. Here, the correlation coefficient when using this equation r _g, when the correlation coefficient r _b when using other formulas, Figure 8 for whether to apply one of the two As described above, four types of combinations are conceivable, and the ◎ portion is effective. Incidentally, r b _-r b _(both r _b) variations in the matching result for variables to enter the formula becomes large in the case of, eliminating the variables by performing level conversion described above with respect to this it can. _Although no particular problem in the results when the r g -r _g, in principle _{r g -r b, r b -r} g, same because the result to be considered to be, the amount of calculation number becomes other in terms It is disadvantageous compared to the combination.
[0040]
Further, as a method of calculating the similarity, for example, there is a method of calculating by calculating the Euclidean distance d between the vectors, and the method is not particularly limited to the method of calculating the correlation coefficient as described in the present embodiment.
[0041]
Further, regardless of whether or not they are adjacent to each other, a plurality of rings may be cut out and the same processing as in the above-described embodiment may be performed. According to this, it is also possible to extract a portion considered to contain important features.
[0042]
【The invention's effect】
As is apparent from the above description, according to the first aspect of the present invention, even if the circular pattern is cut out from any position on the circumference, the correlation coefficient does not change, so that the work of positioning the circular pattern becomes unnecessary. The correlation data can be obtained without being affected by the rotation amount of the circular object. Therefore, the processing time required for identifying a circular pattern can be reduced.
[0043]
The invention of claim 2, wherein, because is the correlation data for one rotation if shifting to half rotate the image data by one pixel, the process less information amount to be when identifying an object, the processing time Can be shortened.
[0044]
According to the third aspect of the present invention, since the circular pattern is composed of a plurality of line data, it is possible to identify an object using a lot of correlation data, and furthermore, the amount of information to be processed Can be reduced.
[0045]
According to the fourth aspect of the invention, it is not necessary to perform the alignment of the circular pattern, and the correlation data can be obtained by shifting the image data by a half turn, so that the template can be easily created. In addition, this can reduce the processing time when a circular pattern is identified using the reference template.
[0046]
According to the fifth aspect of the present invention, it is easy to prepare a template required for identifying a circular pattern, and it is possible to accurately identify a circular pattern of an object using the template.
[0047]
According to the sixth aspect of the present invention, since a pattern characteristically representing a pattern in the classification category is employed as the reference template, a circular pattern as an object can be accurately identified.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an embodiment of the present invention, and illustrates a state in which a part of an image is cut out from a circular pattern in a ring shape and used as image data.
FIG. 2 is a diagram illustrating a state in which image data is shifted one pixel at a time and a correlation coefficient r with original data is obtained for each pixel.
FIG. 3 is a diagram illustrating an example of correlation data indicating a relationship between a calculated correlation coefficient and a shift amount.
FIG. 4 is a diagram illustrating a state in which a part of an image is cut out in a ring shape from a circular pattern and a plurality of lines are extracted as image data.
FIG. 5 is a diagram showing an example of correlation data composed of 10 lines.
FIG. 6 is a diagram illustrating a state in which a reference template is selected from a plurality of sample data.
FIG. 7 is a diagram showing a state of edge enhancement by an edge enhancement filter.
8 is a table showing the effectiveness of the combination of the correlation coefficient r _b and r _g.
[Explanation of symbols]
1 Circular pattern 2 Correlation data 3 Line data 4 Reference template D Image data D ₀ Original image data r Correlation coefficient

Claims (6)

The image data of the circular pattern is read, the read image data is shifted one pixel at a time to obtain a correlation coefficient with the original data, and the correlation data constituted by the correlation coefficient is identified as the identification data of the circular pattern. A method for identifying a circular pattern, wherein the circular pattern is identified using 2. The method according to claim 1, wherein the correlation data is obtained by shifting the image data of the circular pattern by one pixel at a time up to half. 3. The circular pattern according to claim 1, wherein the circular pattern is data of a two-dimensional array configured by collecting a plurality of line data, and the correlation data is obtained for each line data to become comparison data. 4. How to identify the pattern. The image data of the circular pattern is shifted one pixel at a time to obtain a correlation coefficient with the original data. Correlation data composed of the correlation coefficient is prepared as a reference template, and the circular pattern to be compared is prepared. The circular pattern is identified by calculating the correlation coefficient with the original data by shifting the image data one pixel at a time, and comparing the correlation data constituted by the correlation coefficient with the template. A method for identifying a circular pattern, characterized by the following. 5. The method according to claim 4, wherein the reference template is prepared for each type of circular pattern to be classified. 6. The method according to claim 4, wherein a plurality of the correlation data are obtained as a sample, a representative one that accurately indicates the characteristic of the circular pattern is obtained using the correlation data, and the representative data is adopted as the reference template. A method for identifying the circular pattern described.

Images