JP4231476B2 - An image processing method for collating an input image inputted with a printed matter or a sticker label of a container with a master image registered in advance - Google Patents

Description

  The present invention relates to an invention that can be used for inventory management in the distribution industry and quality control in the manufacturing industry.

The correlation coefficient is calculated by coarsening the target image data, and the correlation coefficient is calculated finely at the candidate location.
JP-A-6-160047

  The processing time until collation is completed is faster than that of the conventional technology.

The verification process was carried out by the following four means.
(1) Image input → develop in memory → deletion of unnecessary portion → rough density difference matching → correlation coefficient is calculated and collated.
(2) Image input → develop in memory → delete unnecessary part → matching by rough density comparison → calculate correlation coefficient and collate.
(3) Image input → develop in memory → delete unnecessary part → match by comparison of coarse two-dimensional change amount → correlation coefficient is calculated and collated.
(4) Image input → develop in memory → deletion of unnecessary part → matching by comparison of coarse two-dimensional change amount by binarization → correlation coefficient is calculated and collated.

  The input image obtained by scanning the printed matter or the attached label on the surface of the container is collated with a master image that has been registered in advance, and the processing time until the completion of collation is made faster than the conventional technique and the misidentification rate of collation is lowered, It can be used for inventory management in the distribution industry and quality control in the manufacturing industry.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the collation device of the present invention.

  Image data from a memory 5 for developing image data, an external storage device 6 for storing pre-registered image data (master image), a CRT 4 for displaying results and information, and a container 7 centering on the overall control unit 3 8 (input image) is composed of an input device 1 and an image input unit 2.

(1) While rotating the container 7, the input device 1 scans the surface of the container and takes it into the image input unit 2 of grayscale image data (256 gradations). The overall control unit 3 reads the image data from the image input unit 2 and develops it in the memory 5.
(2) Since the input image developed in the memory 5 includes a non-verification 9 portion as shown in FIG. 1-1, this portion is removed. The external part to be collated is removed by determining that the line is unnecessary when the density values of all the pixels of one line of the input image shown in FIG. Further, the line density of the verification target portion is not determined as unnecessary because a plurality of pixels do not fall within an arbitrary range. By removing lines determined to be unnecessary, an input image in which only the label portion 8 remains is obtained.

  The “method for removing a non-verification part” according to claim 1 is divided into an “input image example” shown in FIG. It explains using.

  In FIG. 1-2, the input image is composed of 40 pixels in the horizontal direction (40 columns) and 23 pixels in the vertical direction (23 lines). [Totaling array for storing totaling results for each line] is a totaling array for storing the results (count values) of density value comparison for each line.

Next, a method for removing the verification target portion (9 shown in FIG. 1-2) will be described with reference to the flowchart shown in FIG. 1-3.
First, the aggregation array that stores the aggregation results for each line is cleared to zero. (Step S1)
1 is set to an index p indicating a line. (Step S2)
1 is set to the index q indicating the column. (Step S3)
It is determined whether the absolute value of the difference between the density value of the q pixel of the p line and the density value of the q + 1 pixel of the p line is 5 or less. (Step S4) The absolute value 5 of the difference is arbitrary.
When the determination result of (Step S4) is Yes, the process proceeds to (Step S5), the count value of the p-line in the total array is incremented by 1, and the process proceeds to (Step S6).
When the determination result of (Step S4) is No, the process proceeds to (Step S6).
In step S6, q indicating a column is incremented by 1, and if q is equal to or less than the number of pixels in the horizontal direction −1, the process returns to step S4. If q is larger than the number of pixels in the horizontal direction −1, the process proceeds to step S8. (Step S7)
In step S8, p indicating a line is incremented by 1, and if p is equal to or smaller than the number of pixels in the vertical direction, the process returns to step S4. If p is larger than the number of pixels in the vertical direction, the process proceeds to step S10. (Step S9)
1 is set to an index p indicating a line. (Step S10)
It is determined whether the count value of the p-line in the aggregate array is 98% or more of the number of pixels in the horizontal direction. (Step S11) This 98% is arbitrary.
When the determination is Yes, the process proceeds to (Step S12). If No, the process proceeds to (Step S13), the p-1 line is removed from the first line of the input image, and the process proceeds to (Step S15). (Step S14)
In (step S12), p is incremented by 1, and if p is smaller than ½ of the number of pixels in the vertical direction, the process returns to (step S11). If p is 1/2 or more of the number of pixels in the vertical direction, the process proceeds to step S15.
The lowest line number is set in the index p indicating the line. (Step S15)
It is determined whether the count value of the p-line in the aggregate array is 98% or more of the number of pixels in the horizontal direction. (Step S16) The 98% is arbitrary.
When the determination is Yes, the process proceeds to (Step S17). If No, the process proceeds to (Step S18), and the p + 1-th line is removed from the lowermost line of the input image, and the process ends.
In step S17, p is decremented by 1. If p is larger than 1/2 of the number of pixels in the vertical direction, the process returns to step S16. If p is ½ or less of the number of pixels in the vertical direction, the process ends. (Step S19)

(3) Next, a “collation method by comparing density differences” according to claim 2 will be described with reference to FIGS. 1-4 and 1-5.
FIG. 1-4 is a “reduced image composed of pixels obtained by extracting image data in 8-pixel jumps”. In this embodiment, 8 pixel jumps are extracted, but the number of pixel jumps is arbitrary. The pixel portion of the oblique draft in FIG.

  1-5 are a [reduced master image S] obtained by reducing the master image by the above-described method and a [reduced input image D] obtained by reducing the input image.

（縮小入力画像Ｄ（ｋ，ｌ）：ｋ＝０，１，２，・・・Ｎ−ｎ、ｌ＝０，１，２，・・・・Ｍ−ｎ）
濃度差分の比較による照合方法は、まず、縮小入力画像Ｄの原点（０,０）から順次、座標をずらしながら縮小マスタ画像Ｓとの濃度差分を算出していき、濃度差分の変化のピークとなる座標(濃度差分の変化が小→大となる座標)を保存する。座標が（Ｎ−ｎ,Ｍ−ｎ）まで進んだら、濃度差分の算出は終了する。
（４）次に保存した座標を一つ取り出し、元の入力画像の座標に変換し、その座標を中心にした縦１６画素、横１６画素の範囲で、マスタ画像と入力画像の相関係数を算出する。なお、この１６画素四方の範囲は任意である。相関係数が最大値となった座標と相関係数を、最も一致した候補として保存する。保存してある残りの座標についても同様の処理を行い、最も一致した候補として保存していく。
（５）次に、最も一致した候補の中から、最大の相関係数を求め、その相関係数の座標をマスタ画像と入力画像の一致した位置とする。 The density difference is calculated as follows.
The density difference between the reduced master image S (size m and width n pixels) and the reduced input image D (size M and size N pixels) is defined as dif.

(Reduced input image D (k, l): k = 0, 1, 2,... N−n, l = 0, 1, 2,... M−n)
The collation method based on the comparison of the density difference first calculates the density difference from the reduced master image S while sequentially shifting the coordinates from the origin (0, 0) of the reduced input image D, (Coordinate where the change in density difference is small → large) is saved. When the coordinates have advanced to (N−n, M−n), the calculation of the density difference ends.
(4) Next, one stored coordinate is taken out and converted to the original input image coordinate, and the correlation coefficient between the master image and the input image is obtained within the range of 16 pixels vertically and 16 pixels horizontally centered on the coordinates. calculate. Note that this 16-pixel square range is arbitrary. The coordinates having the maximum correlation coefficient and the correlation coefficient are stored as the best matching candidates. The same processing is performed for the remaining coordinates that have been stored, and the coordinates are stored as the most consistent candidate.
(5) Next, the maximum correlation coefficient is obtained from the most matched candidates, and the coordinates of the correlation coefficient are set as the positions where the master image and the input image match.

(1) Image data is read from the image input unit 2 and developed in the memory 5. (Same as Example 1)
(2) Since the input image developed in the memory 5 includes a non-verification 9 portion, this portion is removed. (Same as Example 1)
(3) Next, a “collation method by comparing density distributions” described in claim 3 will be described with reference to FIG.
In this embodiment, an image is extracted in 8-pixel jump, but the number of pixel jumps is arbitrary. (Same as Example 1)

(4) FIG. 2-1 shows respective density distributions of the reduced master image S reduced by the above-described method and the reduced input image D reduced the input image.
The density distribution of the reduced input image D is calculated as follows.
When the reduced master image S (size m and width n pixels) and the reduced input image D (size length M and width N pixels), the density of the area of the vertical m pixels and width n pixels in the reduced input image D is shown. Find the distribution.
In the collation method based on the comparison of the density distributions, first, the density distribution is obtained by sequentially shifting the coordinates from the origin (0, 0) of the reduced input image D, and the difference from the density distribution of the reduced master image S is calculated. If a value smaller than the arbitrarily set tolerance (see claim 5) appears, the coordinates are stored. When the coordinates have advanced to (N−n, M−n), the comparison of the density distributions ends.

(5) Here, the processing time is shortened by calculating the feature amount from the difference data and the previously calculated feature amount.
The “method for reducing the number of calculations by calculating only the difference data group” according to claim 4 will be described with reference to FIG.
When the feature amount (density distribution) of each point is obtained from the reduced input image D described above, the feature amount of the next area is calculated as follows.
The previous feature value (density distribution) is Ho, the feature value (density distribution) obtained this time is Hn, the feature value (density distribution) in the front row of the previous feature value (density distribution) Ho is hf, and the feature value obtained this time ( If the feature amount (density distribution) in the last column portion of the density distribution (Hn) is hb, it can be expressed as follows.
Hn = Ho-hf + hb
As shown in FIG. 2B, the feature amount of the next area is calculated from a part of the reduced input image D described above.

例として、表１に入力画像４種類Ａ，Ｂ，Ｃ，Ｄとマスタ画像の照合を行った結果が示してある。例えば候補数が１０件以上存在する許容範囲は、表１から入力画像Ａの場合、６％、Ｂは１０％、Ｃは１１％、Ｄは１０％となる。この許容範囲内の座標群を保存する。 (6) Next, select an allowable range in which the number of coordinates (number of candidates) of the collation result is a certain number or more.
The “selection method of an allowable range in which the number of candidates becomes a certain number or more” according to claim 5 will be described with reference to Table 1.
The method of selecting an allowable range in which the number of coordinates (number of candidates) detected in the matching process between the master image and the input image exceeds a certain number is to detect a group of coordinates in an arbitrary allowable range in advance, and the number of candidates is constant. Select an allowable range that is at least as large as the number.

As an example, Table 1 shows the result of collating the four types of input images A, B, C, and D with the master image. For example, the allowable range in which there are 10 or more candidates is 6%, B is 10%, C is 11%, and D is 10% in the case of the input image A from Table 1. Coordinate groups within this tolerance are saved.

(7) Next, one saved coordinate is taken out and converted to the original input image coordinate, and the correlation coefficient between the master image and the input image is calculated in the range of 16 pixels vertically and 16 pixels horizontally centered on the coordinates. calculate. Note that this 16-pixel square range is arbitrary. The coordinates having the maximum correlation coefficient and the correlation coefficient are stored as the best matching candidates. The same processing is performed for the remaining coordinates that have been stored, and the coordinates are stored as the most consistent candidate. (Same as Example 1)
(8) Next, the maximum correlation coefficient is obtained from the most matched candidates, and the coordinates of the correlation coefficient are set as the positions where the master image and the input image match. (Same as Example 1)

(1) Image data is read from the image input unit 2 and developed in the memory 5. (Same as Example 1)
(2) Since the input image developed in the memory 5 includes a non-verification 9 portion, this portion is removed. (Same as Example 1)
(3) Next, the “collation method based on comparison of two-dimensional density change amounts” described in claim 6 will be described with reference to FIGS.
In this embodiment, an image is extracted in 8-pixel jump, but the number of pixel jumps is arbitrary. (Same as Example 1)

(4) FIG. 3A is a “method for detecting a two-dimensional density change amount of a multi-value image”.
The detection method of the two-dimensional density change amount of the image is performed as follows.
(a) Obtain the average density X and 1/4 of the average density in FIG. This value becomes the threshold value α of the change point.
(b) First, the density difference between pixels adjacent in the horizontal direction is obtained, and when this value is larger than the threshold value α, the number is incremented to obtain the number of change points that are greater than or equal to the threshold value α in one line.
(c) Similarly, in the vertical direction, the number of pixels in which the density difference between adjacent pixels is equal to or greater than the threshold value α is counted for each vertical line.
(d) The number of change points calculated in (b) and (c) is the two-dimensional density change amount.

(5) FIG. 3-2 is a “collation method based on comparison of two-dimensional density change amounts”, which is reduced by the method of FIG. 1-5 [reduced master image S] and reduced input image [reduced input image D]. And the respective two-dimensional density change amounts.
When the reduced master image S (size m, width n pixels) and the reduced input image D (size M, width N pixels) are used, first, the coordinates are sequentially shifted from the origin (0,0) of the reduced input image D. A two-dimensional density change amount is obtained, and a difference from the two-dimensional density change amount of the reduced master image S is calculated, and a value smaller than an allowable range (see claim 5) in which each density change amount difference is arbitrarily set. If it appears, save the coordinates. When the coordinates have advanced to (N−n, M−n), the comparison of the two-dimensional density change amount ends.

(6) The feature amount is calculated from the difference data and the previously calculated feature to shorten the processing time. (Same as Example 2)
(7) Select an allowable range where the coordinates of the matching result are a certain number or more. (Same as Example 2)
(8) Next, one stored coordinate is taken out and converted to the original input image coordinate, and the correlation coefficient between the master image and the input image is calculated in the range of 16 pixels vertically and 16 pixels horizontally centered on the coordinates. calculate. Note that this 16-pixel square range is arbitrary. The coordinates having the maximum correlation coefficient and the correlation coefficient are stored as the best matching candidates. The same processing is performed for the remaining coordinates that have been stored, and the coordinates are stored as the most consistent candidate. (Same as Example 1)
(9) Next, the maximum correlation coefficient is obtained from the most matched candidates, and the coordinates of the correlation coefficient are set as the positions where the master image and the input image match. (Same as Example 1)

(1) Image data is read from the image input unit 2 and developed in the memory 5. (Same as Example 1)
(2) Since the input image developed in the memory 5 includes a non-verification 9 portion, this portion is removed. (Same as Example 1)
(3) Next, a “collation method based on comparison of two-dimensional density change amounts by binarizing an image” described in claim 6 will be described with reference to FIGS.
In this embodiment, an image is extracted in 8-pixel jump, but the number of pixel jumps is arbitrary. (Same as Example 1)

(4) FIG. 4A is a “method for detecting a two-dimensional density change amount of a binary image”.
The detection method of the two-dimensional density change amount of the image is performed as follows.
(a) A threshold is determined from the histogram of the image, and the image is binarized.
(b) Count the number of changes in the density difference between pixels adjacent in the horizontal direction from 0 → 1 or 1 → 0 to obtain the number of change points in one line.
(c) Similarly, in the vertical direction, the change point of the density difference between adjacent pixels is counted from 0 → 1 or 1 → 0 for each vertical line (12 lines).
(d) The number of change points calculated in (b) and (c) is the two-dimensional density change amount.

(5) Collation based on the two-dimensional density change amount is performed. (Same as Example 3)
(6) The feature amount is calculated from the difference data and the previously calculated feature to shorten the processing time. (Same as Example 2)
(7) Select an allowable range where the coordinates of the matching result are a certain number or more. (Same as Example 2)
(8) Next, one stored coordinate is taken out and converted to the original input image coordinate, and the correlation coefficient between the master image and the input image is calculated in the range of 16 pixels vertically and 16 pixels horizontally centered on the coordinates. calculate. Note that this 16-pixel square range is arbitrary. The coordinates having the maximum correlation coefficient and the correlation coefficient are stored as the best matching candidates. The same processing is performed for the remaining coordinates that have been stored, and the coordinates are stored as the most consistent candidate. (Same as Example 1)
(9) Next, the maximum correlation coefficient is obtained from the most matched candidates, and the coordinates of the correlation coefficient are set as the positions where the master image and the input image match. (Same as Example 1)

The block diagram which shows the structure of a collation apparatus. Scanning image of the input object. An example of a scanning image of an input object. The flowchart of the deletion process for the external part for collation. An image obtained by extracting one pixel in 8 pixel jumps. Explanatory drawing of the collation method by comparison of a density difference. Explanatory drawing of the collation method by comparison of density distribution. Explanatory drawing of the calculation method of the feature-value by a difference. Explanatory drawing of the detection method of the two-dimensional density variation | change_quantity of a multi-value image. Explanatory drawing of the collation method by the comparison of a two-dimensional density | concentration change amount. Explanatory drawing of the detection method of the density change amount of a binary image.

Explanation of symbols

1 input device 2 image input unit 3 overall control unit 4 CRT
5 Memory 6 External storage device 7 Container 8 Label 9 Non-verification target 10 Verification target part 11 Total array for storing total results for vertical size

Claims (8)

In the label coincidence determination device that stores the image of the label in advance and determines whether the label attached to the container matches the label of the image stored in advance,
An input device (1) for imaging the container surface to which the label is attached;
When an image of the container surface imaged by the input device (1) is input,
If the density difference between adjacent pixels in each line of the image of the container surface is within a certain value, one is counted per adjacent pixel,
Storing the count for each line;
The total value of the count stored for each line and the number of pixels constituting the line of the container surface image are set to 1 from the extreme end line of the container surface image toward the center of the container surface image. Compare line by line,
A line deleted image is generated by deleting from the line where the difference between the total value and the number of pixels is a certain value or more to one line closer to the endmost line,
Reducing the line-deleted image and the image of the label stored in advance in the external storage device (5) at the same reduction ratio ;
A matching coordinate candidate that is a coordinate candidate whose degree of matching between the label portion in the image of the reduced container surface and the image of the reduced label is equal to or greater than a certain value,
Calculating candidate coordinates corresponding to the matching coordinate candidates in the image of the container surface before reduction ,
An overall control unit (6) for calculating a correlation coefficient with the image of the label at the candidate coordinates in the image of the container surface before the reduction ;
A label matching judgment device comprising: The overall control unit (6)
In the image of the reduced the container surface, and calculating the density difference between the image of the label in the reduced,
The label matching determination apparatus according to claim 1, wherein coordinates having a small density difference are calculated as the matching coordinate candidates. The overall control unit (6)
Calculating a density distribution in the image of the reduced container surface, and a concentration distribution of a label of images the reduced,
The label matching determination apparatus according to claim 1, wherein coordinates having a small difference in density distribution are calculated as the matching coordinate candidates. The overall control unit (6)
Calculating a two-dimensional density variation in the image of the reduced the container surface, and a two-dimensional density change amount of the label of the image the reduced,
The label matching determination apparatus according to claim 1, wherein coordinates with a small difference in the two-dimensional density change amount are calculated as the matching coordinate candidates. The overall control unit (6)
The image of the reduced the container surface, and an image of the label in the reduced binarized,
Calculate the two-dimensional density change amount for each image,
The label matching determination apparatus according to claim 4 , wherein coordinates having a small difference in the two-dimensional density change amount are calculated as the matching coordinate candidates. The overall control unit (6)
In the image of the reduced container surface, when calculating by matching coordinate candidate with the reduced label image, shifting the image of the label with the image of the reduced containers surface the reduced,
The label matching determination apparatus according to claim 1, wherein the matching coordinate candidate is calculated using only the difference when shifted. The overall control unit (6)
When calculating the matching coordinate candidate with the reduced label image in the reduced container surface image,
Wherein as matching coordinate candidate is equal to or greater than the certain number, in the image of the reduced containers surface, wherein the reduced degree of matching labels match according to claim 1, characterized in that to change the difference between the label image Judgment device. In a label matching determination method for storing an image of a label in advance and determining whether a label attached to a container matches the label of the image stored in advance.
The input device (1) images the surface of the container with the label,
The general control unit (6)
When an image of the container surface imaged by the input device (1) is input,
If the density difference between adjacent pixels in each line of the image of the container surface is within a certain value, one is counted per adjacent pixel,
Storing the count for each line;
The total value of the count stored for each line and the number of pixels constituting the line of the container surface image are set to 1 from the extreme end line of the container surface image toward the center of the container surface image. Compare line by line,
A line deleted image is generated by deleting from the line where the difference between the total value and the number of pixels is a certain value or more to one line closer to the endmost line,
Reducing the line-deleted image and the image of the label stored in advance in the external storage device (5) at the same reduction ratio ;
A matching coordinate candidate that is a coordinate candidate whose degree of matching between the label portion in the image of the reduced container surface and the image of the reduced label is equal to or greater than a certain value,
Calculating candidate coordinates corresponding to the matching coordinate candidates in the image of the container surface before reduction ,
A label matching determination method, wherein a correlation coefficient with the label image at the candidate coordinates in the image of the container surface before the reduction is calculated.

Images