JPS6128156B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention, in the field of image processing, relates to a matching device that reads a pre-registered pattern and a target pattern and determines whether the target pattern is the same as the registered pattern. A conventional device of this type is shown in FIG. In the figure, 1 is an image input device, 2 is a binarization circuit that binarizes the output of the image input device 1 and takes it out as a digital quantity, and 3 is a two-dimensional addressable device that stores the output of the binarization circuit 2. an image memory; 4 is a registered pattern memory that stores registered pattern information; 5 is a calculation memory that stores intermediate results of calculations; 6 is a projection histogram calculation unit that calculates a horizontal and vertical projection histogram of the image; 7 is a cross-correlation calculation unit that calculates a cross-correlation coefficient between the projection histograms of the registered pattern and the target pattern, and 8 is a matching unit that determines whether the registered pattern and the target pattern match based on the value of the cross-correlation coefficient. . Next, the operation will be explained. First, an image of a subject is input to the image input device 1. The image input device is
For example, a device like a television camera, which consists of an optical system, a scanning mechanism, and an optical conversion mechanism, scans a spot over the subject in raster mode, converts the density information on the spot into an electrical signal by a photoelectric conversion mechanism, and elutes it. It is. Next, in synchronization with the scanning mechanism of the image input device 1, the binarization circuit 2 compares a preset threshold voltage with the output of the image input device 1 using sample pulses for digitally reading the subject. , the pixels fixed to the object, that is, the pixels on the object, are set to "1", and the background is set to "0".
and sequentially write them into the image memory 3. Reference numeral 4 denotes a registered pattern memory in which a plurality of patterns are registered in advance and has a function of specifying a specific pattern using, for example, a keyboard. Below, the size of the digital image to be handled is M x M.
It will be explained as follows. The projection histogram calculation unit 6 reads out the target pattern and the registered pattern from the image memory 3 and the registered pattern memory 4, obtains horizontal and vertical projection histograms, and stores them in the calculation memory 5. For example, a horizontal projection histogram is a count of the number of pixels fixed on an object for each horizontal line starting from the topmost line of a digital image. Similarly, the vertical projection histogram is
The counts are arranged for each vertical line starting from the leftmost edge. The matching unit 8 calculates the similarity between the target pattern and the registered pattern from the horizontal and vertical projection histograms of the target pattern and the registered pattern, and determines whether the target pattern is the same as the registered pattern based on the similarity of both. Make a judgment. The cross-correlation calculation unit 7 calculates a horizontal or vertical projection histogram from the calculation memory 5 f(i), j=1 to M; target pattern g(j), j=1 to M; registered pattern (g(j)= 0, j<1, orj>M) and the cross-correlation coefficient Λ(τ) seek. The matching unit 8 calculates the similarity S from the output of the cross-correlation calculation unit 7. Find it by For matching both patterns, for example, the horizontal and vertical similarities S _H and S _V are
A match is determined when the condition S _H S _V S _O is satisfied based on a predetermined threshold S _O. Note that the cross-correlation coefficient Λ(τ) has a value of −1Λ(τ)1, and the equality sign of Λ(τ)=1 is f(i) = c・g(i−τ) Λ(τ) The equality sign of =-1 is when f(i) =-c·g(i-τ)c>0 holds true. Therefore, the more similar both patterns are, the closer the value will be to 1. Conventional verification devices are configured as described above, so they can identify fake stamps that are clearly different in only one part, and stamps that are slightly different overall due to the contact of the ink when stamping. , the former should be judged as a mismatch and the latter should be judged as a match, but conventional matching devices have a drawback in that it is difficult to classify the two. This invention was made to eliminate the drawbacks of the conventional ones as described above, and the target pattern is obtained by dividing the projection histogram of one of the registered patterns into several small regions, and for each small region, the projection histogram of the other pattern is divided into several small regions. The object of the present invention is to provide a seal imprint matching device that can classify fakes and noises caused by stamping by determining the cross-correlation coefficient with the projection histogram and determining the degree of similarity from its maximum value and the value of τ at that time. FIG. 2 is a schematic configuration diagram showing an embodiment of the present invention, in which 1 is an image input device; 2 is this image input device 1;
Binarize the output and extract it as a digital quantity 2
digitization circuit, 3 stores the output of the binarization circuit 2;
Dimensionally addressable image memory; 4; registered pattern memory storing registered pattern information; 5;
6 is a projection histogram calculation unit that calculates the horizontal and vertical projection histograms of the image; and 7 is a cross-correlation unit that calculates the cross-correlation coefficient of the projection histograms of the registered pattern and the target pattern. A calculation unit 9 is a similarity calculation unit that calculates the degree of similarity from the cross-correlation coefficient; 8 is a comparison unit that determines whether the two marks match or disagree based on the output value of the similarity calculation unit 9 and outputs the result. be. In FIG. 2, the steps up to obtaining a projection histogram are the same as in the conventional method; an object is inputted by the image input device 1, converted into a binary digital image by the binarization circuit 2, and stored in the image memory 3. As the registered pattern, one of the patterns stored in the registered pattern memory 4 can be designated and read out. The projection histogram calculation unit 6 stores horizontal and vertical projection histograms of the registered pattern and the target pattern in the calculation memory 5. The cross-correlation calculation unit 7 calculates a projection histogram f(i), j=1 to M; target pattern g(j), j=1 to M; registered pattern from the calculation memory 5 (however, g(j)=0, j <1, orj>M) and calculate the cross-correlation coefficient Λ _R (τ) in the specified small region R(k, l). seek. The similarity calculation unit 9 calculates the degree of similarity between the registered pattern and the target pattern, that is, the similarity S from the projection histogram.
For example, the projection histogram of one of the patterns, the target pattern, is divided into several equally spaced small regions R ₁ (k ₁ , l ₁ ), R ₂
(k ₂ , l ₂ )..., Rn (kn, ln). Here, (M is a multiple of n km = M/n (m-1) + 1 lm = M/n・m m = 1, 2, ..., n) Then, in each small area, the cross-correlation calculation unit Find the correlation coefficient ΛRm(τ) and calculate τ appropriately.
The cross-correlation coefficient ΛRm in a small region is
The maximum value X _Rn of (τ) and the deviation τm at that time are Find it by Then, the overall similarity S is determined based on these values. This includes, for example, similarity here

Use [Formula]. If the pattern is locally different, the maximum value X _Rn of the cross-correlation coefficient of a small region including that part will be a low value, or another peak in the vicinity will be found, and X _Rn will be relatively large. However, it happens that the deviation τm at that time deviates from the others. Therefore, the first term in the above equation reflects the case by taking the minimum value of X _Rn in the small region, and the second term reflects τm, m
It is a dispersion value indicating the variation of =1 to n, which reflects the case of , and the constant C is a coefficient indicating the specific gravity of and. The matching unit 8 uses the similarity calculating unit 9 to determine the respective similarities, S _H and S _V from the horizontal and vertical projection histograms, and determines whether both patterns match. and output the result. In this judgment operation, for example, it is determined whether or not the condition S _H・S _V S _O is satisfied using a predetermined threshold value S _O. If the condition is met, both patterns match, and if not, they are mismatched. It is for making judgments. As described above, according to the present invention, when performing matching using a projection histogram, the overall similarity is calculated by dividing the area into small areas and using the most matching point in each small area and the cross-correlation coefficient at that time. This configuration has the effect of making it possible to perform highly accurate matching by classifying fake seal impressions that are obviously different only in parts and genuine seal impressions that are noisy overall.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram showing a conventional seal imprint verification device, and FIG. 2 is a schematic configuration diagram showing a seal imprint verification device according to an embodiment of the present invention. Note that the same reference numerals in the figures indicate the same or corresponding parts. 1: Image input device, 2: Binarization circuit, 3: Image memory, 4: Registered pattern memory, 5: Calculation memory, 6: Projection histogram calculation unit, 7: Cross-correlation calculation unit, 8: Collation unit, 9: Similarity calculation unit.

Claims (1)

[Claims] 1. A hand for inputting seal impressions affixed on documents such as securities as binary digital images, a memory means for storing a large number of binary digital images of seal impressions, and a means for selecting a registered pattern to be compared from this memory means. and determine whether or not the two seal impressions are the same using a projection histogram that counts the number of pixels having a value of "1" in the horizontal and vertical directions for the seal impression on the document and the registered pattern. In the seal imprint matching device, one projection histogram is divided into a plurality of small regions, each small region is matched with the other projection histogram, and the maximum value of the matching degree for each small region and these A seal imprint matching device characterized in that the overall matching degree is determined from the value of the dispersion degree of each maximum value, and based on this, it is determined whether the marks match or do not match.