JPH0573686A - Image inspecting device - Google Patents

Abstract

PURPOSE:To collate two binary images in high speed and to diminish the memory amount of registration information for image inspection. CONSTITUTION:In an image inspecting device 1 checking the first and second image of the binary image, the set of black picture elements is dealt as a line, registration information of the respective picture element address of a first changed image obtained by processing the width of the line of the first image with the fine processing means of line width designation value narrower than the original width of the line is stored in a memory 6, the second image is stored in a memory 4, and in reference to checking the first and second images, the conformity of the first and second images are judged with the collation processing means of registration information of the first image and the respective black picture element address of the second changed image obtained by processing the second image with the narrowing processing means of the line width designation value by a processor 5. In the collation processing means, the enroute abort/decision of an image information moving amount and the two-step-control of that are excuted to reduce a processing amount, and the black picture element of a dissident part is inspected to improve a recognition rate. The processing amount and the memory amount are reduced without decreasing the accuracy of recognition by separation setting of the partial image of positioning.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to pattern recognition of digitized images (fingerprints, imprints, figures, characters, etc.) for image inspection devices (electronic computers, electronic exchanges, communication control devices,
Hardware and / or software in IC cards, image processing devices, image recognition devices, image matching devices, etc.)
The present invention relates to an image inspection device for determining a match or a mismatch based on the matching (degree of matching) of two images by collating the two images.

[0002]

2. Description of the Related Art As an example of an image targeted for pattern recognition, a case where the image is a fingerprint will be described. A fingerprint is a ridge pattern on a finger. Since the valley line (between the ridge lines) is determined by the ridge line, the pattern drawn by the valley line may be used as the fingerprint instead of the ridge line. The line treated as a fingerprint is called a fingerprint line. A fingerprint input device for personal identification is an imaging device (for example, C
As a method for inputting from a CD (charge-coupled device) camera, a prism method (for example, Shimizu et al., “Fingerprint information detection method using prism-comparison between total reflection method and optical path separation method”), IEICE Transactions, Vol. J68-D, No.
3, pp. 414-415 (1985)), and a hologram system (for example, "Personal collation device using holographic fingerprint sensor" by Igaki et al., Technical Report of IEICE, PRU87-31, pp. 27-33, ( 19
1987)) and so on. A fingerprint image of analog information input from the image pickup device is A / D (analog / digital)
The converter converts it into a grayscale image of the digitized fingerprint. The grayscale image of the fingerprint is represented by the coordinates (X, Y) which are the pixel addresses of the image memory and the brightness of the pixels which are the pixel address constituent elements of the image memory. In addition, the unevenness of the fingerprint may be directly converted into a binary image to be a fingerprint image. The grayscale image of the fingerprint can be corrected by smoothing or the direction of the ridge. There are end points, branch points, and intersections as the characteristic points that represent the characteristics of the fingerprint. For the characteristic points of the grayscale image of the digitized fingerprint, the fingerprint image is binarized and further thinned to obtain a range of pixels representing the characteristic points (for example, a set of 3 × 3 pixels centering on the characteristic points). It can be detected by the presence of the same pattern as the pattern in the thinned image (eg, Sasakawa et al., "Fingerprint collation device for personal identification with improved ability to cope with low quality images", IEICE Transactions, Vol.J72. -D-II, No. 5, pp. 707-714
(1990)).

In collating fingerprints, a fingerprint in which information for collating is registered in a memory in advance is referred to as a registered fingerprint, and a fingerprint whose matching with the registered fingerprint is collated is referred to as an inspection fingerprint. Known methods of collating the registered fingerprint and the inspection fingerprint include a method using a feature point, a method using the direction of a ridge, and a method using pattern matching between original images of the registered fingerprint and the inspection fingerprint. As a method of pattern matching between thinned images, Japanese Patent Laid-Open No. 63-132386 describes a matching method by superimposing a thinned image of a registered fingerprint and a thinned image of an inspection fingerprint.

Smoothing is a process for reducing the influence of noise on a fingerprint image. For example, there is a locally weighted average filter that uses the values of neighboring pixels of an arbitrary pixel. Takagi and Shimoda (supervised) "Image Analysis Handbook" , Pp. 538-548,
The University of Tokyo Press (1991), etc. For thinning a binary image, the line width of most of the pixels (the majority means more than half to all, and ideally all) of the pixels of the target line type is one pixel. The target type pixel is either a black pixel or a white pixel, but will be described as a black pixel hereinafter. As a method of binarizing a grayscale image and thinning the binary image,
Within the set of black pixels, there is a Hilditch thinning method in which outer black pixels are sequentially deleted while retaining the black pixel connectivity (4 connections or 8 connections) (for example, Tamura (Supervised) "Introduction to Computer Image Processing", Soken Shuppan, pp.80-83 (1985), Tamura, "Study on Multifaceted Image Processing and Its Software System", Research Report of Electrotechnical Laboratory, pp. 25-
64,835 (February 1984), and Mori et al., "Basics of Image Recognition [I]," pp. 65-71, Ohmsha (1
986)). Kobayashi, "Image thinning and feature point extraction",
IEICE Technical Report, PRU 90-149, pp. 33-38 (19
1991) also describes a thinning method for a grayscale image or a binary image.

A method of binarizing a grayscale image into a binary image is described in, for example, "Basics of Image Recognition [I]" by Mori et al.
37-47, Ohmsha (1986), and the like. When the fingerprint is input, the inspection fingerprint and the registered fingerprint are misaligned (rotated and / or moved in parallel). Therefore, the verification of the inspection fingerprint and the registered fingerprint requires alignment of both fingerprints. As the alignment method (rotation and parallel movement), the method that uses the ridge direction, the method that uses the representative feature points and the peripheral feature points, and the trial and error for the range in which only parallel movement is possible Positioning method and the like are known. The formulas for coordinate transformation and geometric transformation required for alignment are described in, for example, Plastock et al., Koriyama Translation “Computer Graphics”, pp. 84
-88, McGraw-Hill Book (1987).

It is useful to find the approximate center point of the fingerprint image in the alignment at the time of matching. Japanese Patent Office Sho 5
8-55548 "Graphic center position determination method" describes a method of sequentially searching a direction in which the slope of a ridge is steep to obtain a center point. Ito et al., "Classification Method of Fingerprint Images Focusing on Central Point," IEICE Tech., PRU 89-79, pp. 1
5-22 (1989) describes a method of sequentially approaching the center position by using the number of intersections with parallel lines on each side of a rectangle. In "A Study on Extraction of Reference Points in Fingerprint Matching", 1987 IEICE National Conference on Information and Systems, No. 125, the number of ridges passing through each scanning line was counted and the distribution of the number of lines was counted. Are seeking.

Kobayashi, "Consideration of Fingerprint Image Matching Method," IEICE Technical Report, PRU91-45, IEICE, pp. Two
5-30 (July 1991), template matching of a black image obtained from a thinned image of a registered fingerprint (or an image subjected to the thinning process) and a binary image of the inspection fingerprint (or an image subjected to the thinning process). Has been proposed, and in this method, the processing amount and memory amount are reduced as compared with the method of performing template matching between binary images.

[0008]

The method proposed by Kobayashi in "Documentation of Fingerprint Image Matching Method" in the above document has the following problems. Problem: Since there are many movement patterns associated with movement during verification, a lot of processing time is required.

Problem: Since the sub-template portion is a continuous area, the processing time at the time of matching is long. The file size of registration information for storing sub-templates and / or non-sub-templates is large. Problem: False recognition occurs because the matching rate of matching increases even when there are many mismatched portions.

Problem: The center point of the fingerprint image may not exist or may be unclear, and the matching process in that case may not be successful. Although not mentioned in the above-mentioned document, the following problems need to be solved in order to realize collation of fingerprint images. Problem: In order to determine the effective area of the fingerprint image, it is necessary to check the distribution of brightness of the grayscale image, which increases the processing time.

Problem: Since the trigger for fixing the image in the image memory by the image inspecting apparatus is unknown, the user needs to instruct the image inspecting apparatus for the trigger, resulting in poor operability.

[0012]

The problems to be solved by the invention will be solved by the following means. Note that these means may be independently selected and used. For the above-mentioned problem, means for abandoning processing midway during collation and / or means for midpoint determination during collation for any specified amount of change in image information movement (rotation, parallel movement) and / or image This is solved by means of changing the step size (change amount) of information movement in two steps. (Refer to claims 1, 2, and 3.) The above problem is solved by means for checking the black pixel in the non-matching portion when the matching rate is highest. (Refer to claim 4.) For the above problems,
The sub-template portion and the non-sub-template portion are solved by means for creating each in an arbitrary number of separation regions. (Refer to claims 5 and 6.) For the above problems,
This is solved by means of providing a range limitation of the approximate center point.
(Refer to claim 7.) The above problem is solved by means for judging the distribution state of the luminance of the binary image. (Refer to claim 8.) The above problem is solved by means of performing image input by checking the effective partial area in the small area (small window area) of the image for determining the image input. (See claim 9.)

[0013]

The means (1) to (3) in the means for solving the problems have the following functions. The unnecessary collation can be abandoned on the way by the collation midway abandonment means and / or the midway determination means, so that the processing amount can be reduced. In the two-step matching means, the moving amount of the image information is set in two steps, so that the matching is performed by moving the image information with a relatively large moving amount in a relatively wide range without lowering the recognition accuracy. The collation can be performed in one stage and the second stage in which the image information is collated by moving the image information with a relatively small movement amount in a relatively narrow range.
Therefore, it is possible to reduce the entire processing amount as compared with the case where one image information is moved with a small moving amount in the entire range.

By performing the check of the non-coincidence portion in the above-mentioned means together with the check of the coincidence portion, there is an effect that erroneous recognition is reduced. By separating and setting a plurality of sub-template parts in the above means, it is possible to reduce the amount of collation processing as compared with the case where continuous sub-template parts are used without lowering the recognition accuracy. Further, there is an effect that the file amount can be reduced by separating the non-sub-template portions and setting a plurality of them.

By limiting the range of the approximate center point in the above means, it is possible to easily align the fingerprint image with a clear center point and to collate even a fingerprint image with an unclear center point. There is an effect that there is. There is an effect that the fingerprint boundary area can be easily determined by the determination of the effective area by the above means.

The image inspection apparatus has the effect of easily inputting an image of good image quality when the image input is confirmed by the above means.

[0017]

EXAMPLES As an example, a case where an image is a fingerprint (sometimes called a fingerprint image) will be described. FIG. 1 is an example of a fingerprint identification system. The fingerprint input from the image input device 2 is processed in the image inspection device 1. The image inspection apparatus 1 includes an image memory 4 for storing a digitized grayscale image of a fingerprint, a binary image, and an image subjected to various processes (thinning process, etc.) when necessary, a processor (one or more). It is provided with a processing device 5 which is a CPU) and a memory 6 for storing information such as programs, data, files (collection of data). A storage device having different characteristics in the memory 6 (for example, a semiconductor memory and a magnetic disk)
When there is a mixture of the two, the movement of information between them is performed by hardware or software as needed. The image memory 4 and the memory 6 are divisions based on stored information, and can be realized by the same storage device.
The image input device 2 includes an image pickup device 7. The A / D converter 3 converts analog information into digital information (here, if an image input device of a type that can directly obtain a digital image is used, the A / D converter is unnecessary). The pixel address in the image memory 4 that stores the fingerprint image, which is a grayscale image of the digitized fingerprint, is X
It is expressed as (X, Y) by the coordinates and the Y coordinate. Pixel address (X, Y) is the pixel (X, Y), or simply
Sometimes expressed as (X, Y). A portion of the image memory 4 that stores one image is called a screen. The image memory 4 can have one or more screens.

Each screen of the image memory 4 is composed of pixels, and the range of all pixel addresses is _0≤X≤Xh , 0.
If ≦ Y ≦ Y _h , the processing range further specified within this pixel address range is to be processed. If a number after the decimal point is generated in the pixel address or the number of pixels due to the calculation including the pixel address or the number of pixels, the process is performed by rounding down, rounding down or rounding up. Pixel values are represented by brightness. Which part of the luminance becomes the ridge depends on the processing method of the image input device 2 and the image processing in the image inspection device 1, and in any case, the characteristic of the luminance corresponding to the ridge is determined by the image inspection device 1. It is possible to process by setting in advance. A set of one or more pixels is called a pixel set. In the identification of the fingerprint, the image input device 2 for registering in the memory 6 of the image inspection device 1
The fingerprint input from is called a registered fingerprint, and the fingerprint input from the image input device 2 for inspection is called an inspection fingerprint. In an image binarized into black pixels and white pixels, the fingerprint line is
Either black pixels or white pixels can be selected (any one of them may correspond to a ridge line and a valley line, respectively).

In this embodiment, black pixels are treated as fingerprint lines. The thinning is a process in which most of the line width is set to one pixel, but in the present embodiment, the line width of the black pixel set is included in the image of black pixels of the original binary image. Is called a thinning process, and an image obtained as a result of the thinning process is called a thinning image. Therefore,
Thinning is one form of thinning processing in this embodiment. In addition, line width, when the point of the edge of an arbitrary line is determined,
It is defined as the minimum distance (the number of pixels) of reaching the other edge through the line (determined by the position of the edge of the line).

FIG. 2A shows the state of the image data stored in the image memory 4. The screen of image 10
An image (binary image or grayscale image) obtained from an image input from the image input device 2 and digitized is stored. For example, a past image is stored on the screen of the image 11 and can be used for processing the image 10. The image 11 may be unnecessary depending on the selection of the processing means. In that case, the image memory 4 may be the image 10 only. Figure 2 (b) shows
The situation of data and the like stored in the memory 6 is shown, the program and data 12 store the program and data for realizing the present embodiment, and the registration information 13 contains
Store the registration information of the registered fingerprint image in a file.

When the image is a grayscale image, a code value corresponding to the luminance is set. The binary image may be obtained by binarizing the grayscale image set in the image memory 4 or may be directly set in the image memory 4 depending on the image input method. A binary image is represented by only black and white pixels, but code values corresponding to the respective luminance values of the black and white pixels (for example, 1 for black pixels and 0 for white pixels).
Shall be specified. Whether the black pixel has a high brightness part or a low brightness part depends on the target image and the input method thereof, and may correspond to either case. The logical origin and coordinate axes of the image stored in the image memory 4 can be determined independently of the physical pixel position of the image memory 4. The X axis and Y axis can be set freely, but for convenience of explanation, the X direction is from left to right (that is,
The X direction is the horizontal direction, and the Y direction is the vertical direction from top to bottom (that is, the Y increasing direction).

Notations and definitions are given below, but some of these notations may not appear in the description below.

[0023]

[Equation 1]

F (X, Y): pixel address (X, Y)
Is the brightness at. FA: Fingerprint area. Since the fingerprint area can be expressed only by the fingerprint boundary, it is stored in the memory 6 as fingerprint boundary information. The fingerprint boundary is treated as within the fingerprint area. (X _C , Y _C ): Address of the approximate center point of the fingerprint. (X _RC , Y _RC ): Address of the approximate center point of the registered fingerprint.

(X _TC , Y _TC ): Address of the approximate center point of the inspection fingerprint. Rth: Registered fingerprint changed image (that is, first changed image) obtained by at least thinning the registered fingerprint binary image (first image) obtained from the registered fingerprint image (grayscale image or binary image) Is. Tth: An inspection fingerprint change image (that is, a second image obtained by performing at least thinning processing on a registered fingerprint binary image obtained from the inspection fingerprint image (grayscale image or binary image).
(Changed image).

RA: A set of addresses (X, Y) of all black pixels in the fingerprint area in the registered fingerprint change image Rth. RA is the union of RT (0), RB (0), and unused fingerprint area black pixels. RT (0): RT (0) is a subset of RA. R
T (0) can specify an area of one or more arbitrary separated pixel sets as a small area of RA. RT (0) is called a sub template. The part of the image from which the sub-template is extracted may be called the sub-template part.
This sub-template is used for image registration. (See FIG. 3.) RB (0): Address set of black pixels in the portion excluding RT (0) from RA. RB (0) is a subset of RA. Any number of one or more separate regions of the pixel set can be specified. RB (0) is called a non-subtemplate. The part of the image from which the non-subtemplate is extracted may be called the non-subtemplate part. Non-subtemplates are not used for image registration. (See FIG. 3.) RT (S): Sub-template R when the coordinate axis of the registered fingerprint change image is rotated S degrees about an arbitrarily determined point.
It is a set of addresses (X, Y) of black pixels of T (0).

RT (S, H, V): The coordinate axis of the registered fingerprint change image is rotated S degrees about an arbitrary point (for example, the approximate center point of the registered fingerprint), and then horizontal movement H and vertical movement V are performed. It is a set of addresses of the sub-template RT (0) on the coordinate axis after the execution. RT (0,0,0) = RT
(0), RT (S, 0, 0) = RT (S). (One pattern in which the registered fingerprint change image is moved by coordinate conversion is determined by a set of values of S, H, and V that specify the moving position.) RB (S, H, V): Coordinate axis of the registered fingerprint change image S around an arbitrary point (for example, the approximate center point of the registered fingerprint)
It is a set of addresses of the non-sub-template RB (0) on the coordinate axis after the horizontal movement H and the vertical movement V after the rotation. RB (0,0,0) = RB (0), RB
(S, 0,0) = RB (S).

N1m is the number of sub-template matching black pixels, and represents the number of matching black pixels between the sub-template black pixel of the registered fingerprint and the inspection fingerprint black pixel. N1c: The total number of black pixels of the sub template, which represents the total number of black pixels of the sub template of the registered fingerprint. N2m: The number of non-sub-template matching black pixels, which represents the number of matching black pixels between the non-sub-template black pixels of the registered fingerprint and the inspection fingerprint black pixel.

N2c: The total number of black pixels of the non-subtemplate, which represents the total number of black pixels of the non-subtemplate of the registered fingerprint. Nm: N1m or N2m. Counter Nm is the value of the counter for the calculation of Nm. Nc: N1c or N2c. Counter Nc is the value of the counter for the calculation of Nc. 3 and 4 show, in the image 10 in the image memory 4, the fingerprint area defined by the fingerprint boundary and the portion and the image of the sub template RT (0) that is a subset of the image for performing the image alignment. An example of the relationship of the parts for extracting the non-sub-template RB (0), which is a subset of images other than the subset of images for performing alignment, is shown. Example 1 is a case where the sub-template part and the non-sub-template part are connected and divided,
Although it is known, it can be used in the present invention. Examples 2 to 4
This is an example of division according to the present invention. In Example 2, a plurality of sub-templates RT (0) and non-sub-templates RB (0) are provided, and these are separated from each other by pixel sets other than sub-templates and non-sub-templates. ing. In Example 3, the number of non-sub-templates is continuous and one sub-template RT (0) is separated. Conversely, a plurality of non-sub-templates RB (0) may be separated by one continuous sub-template RT (0). In Example 4, a non-sub-template RB (0) is provided to surround the sub-template RT (0) apart from the sub-template RT (0).

The means for processing the fingerprint image will be described below. If there is no description of the process to be performed next in a step in the procedure of each means, the process immediately proceeds to the next step. The constant used in each means is the image inspection device 1
In addition, in consideration of various conditions in the image input device 2, an appropriate value is set statically or dynamically. The implementation of each step can be modified as long as the content of the process is the same. (1) Smoothing processing means Smoothing is processing for reducing the influence of noise in the fingerprint image. A known method can be used for the smoothing process (for example, a local weighted average filter using the values of neighboring pixels of an arbitrary pixel can be used).

When the binary image can be directly input to the image memory 4, the smoothing is omitted. (2) Binarization and Background Separation Processing Means Binarization is a process of converting a grayscale image into a binary image.
Background separation is a process of clarifying the effective range of the fingerprint image of the image 10 in the image memory 4. An example of a procedure for performing binarization and background separation will be described in procedure B. The input information of the procedure B is input image information. The output information of the procedure B is the binary image of the output image and the fingerprint boundary information. Image memory 4
When an image input device of a type capable of directly inputting a binary image is used, the binarization (step B1) is omitted and only background separation is executed. (Procedure B) Let K be a constant representing the length of the partial region in the X direction (= the length of the partial region in the Y direction).

Let L = (X _h +1) / K. Here, K is selected so that L is an integer (generally, the length of the partial area can be made variable for each partial area). Let K _max = ((X _h +1) / K) −1. For the image 10, the partial area address for uniquely identifying the partial area is set as the leading pixel address of each partial area. An arbitrary pixel address (X, Y) is M = [X / K] and N at the partial area address J (M, N).
= [Y / K]. Therefore, the partial area address J
The pixel address (X, Y) corresponding to (M, N) is: X = K · M, (M = 0, 1, 2, ..., K _max ) Y = K · M, (N = 0, 1, 2, ..., K _max ). Step B1 (binary conversion): M = 0,1,2, ..., K max N = 0,1,2, ..., the K _max, the following process is performed. Each partial area J (M, N)
, The average brightness value Bav of all pixels in each partial area
(M, N) is calculated by Bav (M, N) = (sum of brightness of pixels in partial area) / (number of pixels in partial area). Next, (M, N) in X = K · M, (M = 0 to K _max ) Y = K · N, (N = 0 to K _max ) covers the range of the image 10.
That is, when the image range is 0 ≦ X ≦ Xh 0 ≦ Y ≦ Yh, the range of M = 0 to [Xh / K] N = 0 to [Yh / K] is targeted. At this time, the total number of partial areas of the image 10 = ([Xh / K] +1).
([Yh / K] +1). B for each subregion
For the brightness f (X, Y) of the pixel (X, Y) of the partial area for which av (X, Y) is obtained, a threshold value T for binarization is set as follows: T = average brightness of partial area + D (D is Is a constant of integer (positive, negative, or zero), and (X,
Y), that is, for (X, Y) in the range of K · M ≦ X ≦ K · (M + 1) −1 K · N ≦ Y ≦ K · (N + 1) −1, f (X, Y) ≧ T At this time, (X, Y) is set to a white pixel (however, when inverting black and white, it is set to a black pixel). When f (X, Y) <T, (X, Y) is set to a black pixel (however, when black and white is inverted, it is set to a white pixel).

Here, B ₁ and B _H are constants for distinguishing the effective partial area and the invalid partial area. Step B2 (determination of effective partial area): (Invention of claim 8) For the binary image obtained as a result of Step B1,
The partial area brightness average value is obtained again. Next, partial area address I _MN = ([X / K], [Y / K]) (the partial area address is the start address of each partial area) X = K · M, (M = 0, 1, 2, ..., K _max ) Y = K · N, (N = 0,1,2, ..., K _max ), each of which is calculated by the brightness average value Bav (X, Y) obtained by the partial area brightness average value function. Effective / for each partial area
Determine invalid. That is, when _BL ≤ Bav (X, Y) ≤ _BH , it is determined that the partial area is the effective partial area, and the effective display is set in the effective partial area table Y (M, N). Here, _BL and _BH are constants for distinguishing the effective partial area and the invalid partial area.

[0034]

[Equation 2]

The invalid partial area of the image 10 is set to a white pixel. To obtain the effective partial area directly from the grayscale image, it is necessary to check the luminance distribution for each area, and the processing becomes complicated. Therefore, once binarized, the average luminance is obtained from the binarized image and processed. Step B3 (inspection of the number of effective partial areas): Y (M,
According to N), YT = the number of effective partial areas is counted, and if YT ≧ YC (YC is a constant), then go to step B4.

If YT <YC, this procedure returns with an error due to an error of the number of effective partial areas. Step B4 (left end of fingerprint boundary in unit of partial area): partial area fingerprint boundary information is {(N _T , M _L , M _R ), N _T = 0
˜K _max }, and for each N = N _T value, M _L ≦ M
≦ M _R is a fingerprint area in sub-area units.

Here, the left end M _{L of the} fingerprint boundary is obtained for each partial area. Starting from N = 0, N = N _T , (N
_The following processing is performed for _T = 0 to K _max ). Left edge (M
= 0) to M in the increasing direction, 1 (effective partial area display) of the element G (M, N) of the effective partial area table G is sequentially searched, and 1 is K _C or more (K _C is a partial area). The value M _L of the first M of a continuous portion for determining the fingerprint boundary of is the left end of the fingerprint area for N at that time. When 1 (effective area display) cannot be found continuously for K _C or more at M = 0 to K _max , the value of N is a non-fingerprint area for all M, and N _{T of the} non-fingerprint area is _Let M _L = M _R = −1. Step B5 (right edge of fingerprint boundary for each partial area): Here, the right edge M _{R of the} fingerprint boundary is obtained for each partial area.
Starting from N = 0, N = N _T , (N _T = 0 to K _max ).
, The left end is not set in step B4
(That is, N when M _L = −1) is skipped, and 1 (effective) of the element G (M, N) of the effective partial area table G is sequentially obtained in the decreasing direction of M from the right end (M = K _max ). The partial area display) is searched for, and the first value M _R of the portion where 1 is K _C or more (K _C is a constant) is set as the right end of the fingerprint area for N at that time. From the above, the partial area fingerprint boundary information {(N _T , M _L , M _R ), N _T = 0 to K _max } is obtained. Step B6 (fingerprint boundary information for each pixel): Partial area fingerprint boundary information {(N _T , M _L , M _R ), N _T = 0 to K
The fingerprint boundary information for each pixel is obtained from _max }. Fingerprint boundary information for each pixel is expressed as {(Y _T , _XL , X _R ), Y _T = 0.
Expressed in to Y _h -1}, which is the value of each Y _T, X
_L ≦ X ≦ X _R means the fingerprint area.

Depending on the partial area fingerprint boundary information, N = 0 to K
_{At max} , for Y = Y _T of K · N ≦ Y ≦ K · N + K _max , the fingerprint boundary information for each pixel {(Y _T , K _L = K · M _L X _R = K · M _R + K _{max XL} , X
_R ), Y _T = 0 to Y _h -1}. (End of procedure B) Note that, hereinafter, when simply referred to as fingerprint boundary information, fingerprint boundary information for each pixel {(Y _T , _XL , X _R ), Y _T =
Means 0~Y _h -1}. (3) Setting means for triggering input confirmation from the image input device to the image memory The image 10 input from the image input device 2 to the image memory 4 changes due to the movement of the object of the image pickup device 7, and therefore the image memory 4 It is necessary to determine an opportunity to determine the image 10. To this end, the image input apparatus 2 notifies the image inspection apparatus 1 of a signal confirmation trigger by a signal, the user instructs the image inspection apparatus 1 to confirm the image confirmation trigger, or There is a process in which the image inspection device 1 determines the timing of image confirmation by checking the state of the image 10 in the image memory 4. Here, an example of the processing of is shown.

The trigger for image input is determined by the effective partial area ratio of the small area of the image memory 4. The effective partial area ratio is
It can be obtained in the same manner as in step B1 to step B3 of procedure B. The procedure is shown in procedure Bs (the value of the constant in procedure Bs may be different from the value of the constant in procedure B). When the ratio of the effective partial area is equal to or smaller than the threshold value by the procedure Bs, the procedure Bs is retried after waiting for a fixed time. The input information of the procedure Bs is the input image information and the range of the small window area in the image memory 4. The small window area is a range included in the range of the image 10 used in the procedure B. The reason why the processing range is smaller than that in the procedure B is to shorten the processing time. If the processing time does not matter, all the images 10 used in the procedure B may be targeted. The output information of the procedure Bs is whether or not the image 10 can be confirmed. (Procedure Bs) (Corresponding to the Invention of Claim 9) Let K be a constant representing the length of the partial region in the X direction (= the length of the partial region in the Y direction).

Let L = (X _h +1) / K. Here, K is selected so that L is an integer (generally, the length of the partial area can be made variable for each partial area). Let K _max = ((X _h +1) / K) −1. For the image 10, the partial area address for uniquely identifying the partial area is set as the leading pixel address of each partial area. An arbitrary pixel address (X, Y) is M = [X / K] and N at the partial area address J (M, N).
= [Y / K]. Therefore, the pixel address (X, Y) corresponding to the partial area address J (M, N) is: X = K · M, (M = 0, 1, 2, ..., K _max ) Y = K · N, ( N = 0, 1, 2, ..., K _max ). Step Bs 1 (binarization): M = 0,1,2, ..., K max N = 0,1,2, ..., the K _max, the following process is performed. Each partial area J (M, N)
, The average brightness value Bav of all pixels in each partial area
(M, N), Bav (M, N) = (sum of brightness of pixels in partial area) /
(Number of pixels in partial area)

Next, at X = K · M (M = 0 to K _max ) and Y = K · N (N = 0 to K _max ). (M, N) targets only the small window area in the following steps. That is, when the small window area is X _S ≦ X ≦ X _S + _XL −1 Y _S ≦ Y ≦ Y _S + Y _L −1, M = [X _S / K] to [(X _S + X _L −1 ) / K] N = [Y S / K] to target range _{~ [(Y S + Y L} -1) / K].

Total number of partial areas in the small window area = ([(X _S + _XL −1) / K] − [X _S / K] +1) ·
([(Y _S + Y _L −1) / K] − [Y _S / K] +1) For each partial region, the brightness f (X) of the pixel (X, Y) of the partial region for which Bav (X, Y) is obtained , Y), the binarization threshold value T is T = average brightness of partial area + D (D is an integer (positive, negative, or zero) constant), and the partial area at this time ( (M, N) for (X,
Y), that is, for (X, Y) in the range of K · M ≦ X ≦ K · (M + 1) −1 K · N ≦ Y ≦ K · (N + 1) −1, f (X, Y) ≧ T At this time, (X, Y) is set to a white pixel (black pixel when black-and-white inversion is designated). When f (X, Y) <T, (X, Y) is set to a black pixel (white pixel when black-and-white inversion is designated). Step Bs 2 (determination of effective partial area): Step Bs
For the binary image obtained from the result of 1, the partial area luminance average value is obtained again. next,

[0043]

[Equation 3]

(Partial area address is the start address of each partial area) X = K · M, (M = 0, 1, 2, ..., K _max ) Y = K · N, (N = 0, 1) , 2, ..., K _max ) in the small window area, by the average brightness value Bav (X, Y) obtained by the partial area average brightness function,
Validity / invalidity is determined for each partial area. That is, when _BL ≤ Bav (X, Y) ≤ _BH , it is determined that the partial area is the effective area, and the effective display is set in the effective partial area table Y (M, N). Here, _BL and _BH are constants for distinguishing the effective partial area and the invalid partial area.

[0045]

[Equation 4]

Step Bs 3 (inspection of the number of effective partial areas): Y (M, N) counts YT = the number of effective partial areas in the small window area, and YT ≧ YCB, (YCB is a constant. ), The image input can be confirmed and the process ends.

When YT <YCB, the image input cannot be determined and the process ends. (End of procedure Bs) (4) Processing means for image correction It is necessary to determine how much image correction is performed depending on the image quality, and if necessary, omit or enhance. For example, for the image 10, the isolated black pixel set is removed (changed to white pixels) and the isolated white pixel set is filled (changed to black pixels) by the pixel pattern around each pixel in the fingerprint area of the binary image. It is possible to perform simple image correction such as. (5) Processing means for obtaining approximate center point The procedure for obtaining the approximate center point (X _C , Y _C ) of the fingerprint area from the binary image of the fingerprint in the image 10 is shown in the following procedure Q. The input information of the procedure Q is input image information and fingerprint boundary information. The output information of the procedure Q is the approximate center point (X _C , Y _C ). FIG. 7 shows the relationship between each constant, the fingerprint area, and the approximate center point. (Step Q) step Q1 (processing for calculating a Y _C): for the range of V _L ≦ X ≦ V _R, the horizontal line _{Y = KA i, (i =} 1,2, ..., n; H D ≦ Y
≦ H _U ) and the number of intersecting lines KH _i of the black pixel line which is the fingerprint line,
(I = 1, 2, ..., N) is calculated. Here, the line segment width of the black pixel in each intersecting portion is set to a certain value (for example, one pixel) or more. _{H D, H U, V L} , V R is, V _L ≦ X ≦
V _R, and H _D ≦ Y ≦ H _U sections is the fingerprint area, and a selected constant such that the setting range of the approximate center point KA _{i, (i} = 1,2, …, N) is
It is a constant that determines the scanning position.

Next, the number of intersecting lines K with respect to the horizontal line Y = KA _i
For H _i , (i = 1, 2, ..., N), KH _i , (i
= 1, 2, ..., N) is maximized to obtain i = i _m . Generally i _m is (the T _y pieces) are one or more. Y _C , which is the Y coordinate of the approximate center point,

[0049]

[Equation 5]

Step Q2 (Process for obtaining X _C ): Vertical line X = KB in the range of H _D ≦ Y ≦ H _U
i , (i = 1, 2, ..., N; V _L ≦ X ≦ V _R ) and the number of intersection lines KV _i , (i = 1, 2,
..., n) is calculated. Here, the line segment width of the black pixel in the intersecting portion is set to a certain value (for example, one pixel) or more. KB _i ,
(I = 1, 2, ..., N) is a constant.

Next, the number of intersecting lines K with respect to the vertical line X = KB _i
For V _i , (i = 1, 2, ..., N), KV _i , (i
= 1, 2, ..., N) is maximized to find i = i _s . In general, there are one or more i _s (T _x pieces). X _C , which is the X coordinate of the approximate center point,

[0052]

[Equation 6]

Step Q3 : (Corresponding to the invention of claim 7) (X _C , Y _C ) is set as an approximate center point of the fingerprint area.
(End of procedure Q) Limits are set on the range that the approximate center point can take. The means are as follows. The range of the approximate center point can be set to any shape, and is set to a small range that can be dealt with by the matching process.
For example, it may be in the range of X _CL ≤X ≤X _CH and Y _CL ≤Y≤Y _CH . Outside this range (X _C ,
When Y _C ) is obtained, the approximate center is set to the closest value in the approximate center range. That is, when X _C <X _CL , X _C = X _CL , and when X _C > X _CH , X _C = X _CH , Y _C
<When the _{Y CL Y C = Y CL,} Y C> Y when the _CH Y _C =
Y _CH

As an alternative to the procedure Q, the image 1
0 center ([Xh / 2], [ Yh / 2]) is means be regarded or approximate center point a point near (X _C, Y _C) and. (6) Thinning processing unit A set of one or more black pixels is treated as an image line. The method of holding the line width designation value is arbitrary, and it is possible to use it as the input information of the thinning processing means or to hold it in the thinning processing means. The line width designation value is 1 in one thinning processing unit.
You may keep more than one. In the present embodiment, the thinning means can be selectively used according to the line width designation value. However, the image inspection apparatus 1 only needs to include thinning processing means for the line width designation value applied to the first image and thinning processing means applied to the second image. An example of the thinning processing means depending on the specified line width value is shown below.

Thinning processing for making most of the line width 1 pixel (line width specified value is 1 pixel) Means for thinning the line width of a binary image (or means for binarizing a grayscale image and narrowing the line width) It is possible to use a known method (for example, a thinning method capable of making most of the line width one pixel). There is also a method of directly binarizing and thinning a grayscale image.

Thinning processing for thinning most of the lines to a value equal to or less than the designated value of the line width (the designated value of the line width is an arbitrary value) This can be realized based on a known method. For example, there are the following means. a) Most of the line width becomes 1 pixel by the processing for one screen in which line elements are deleted pixel by pixel from the outside of the line (corresponding to the black pixel which is the fingerprint line in the present embodiment) forming the image. In the case of the thinning method that iterates up to, in the thinning process, the number of iterations = {(the maximum line width of the original binary image)-(line width specified value)} / (each line width in the deletion process for one screen. It is sufficient to set the number of iterations so that it becomes an approximate value of the number of pixels deleted each time).

B) In the case of the thinning method in which a black pixel is left from the center of the element of the line forming the image, the center of the line segment is included for the black pixel on the line segment where the line of the image intersects the line of the image. Pixels less than the line width specified value are left (when the line segment width is greater than or equal to the line width specified value, black pixels for the line width specified value are left centered at the midpoint of the line segment and the line segment width is less than the line width specified value This can be achieved by leaving all black pixels on the line segment.

In the present embodiment, the line width designation value for performing the thinning process on the registered fingerprint image and the inspection fingerprint image can be arbitrarily designated, but the quality and characteristics of the line of the input image and the line width of the thinning process can be specified. It is necessary to determine it from the performance with respect to the specified value, the performance required for the image inspection apparatus 1, and the like. Increasing the difference between the line width designation values in the thinning processing for the registered fingerprint image and the inspection fingerprint image will strengthen the positional deviation between the two images, but if the difference is too large, the matching accuracy may decrease. is there. The smaller the line width of the registered fingerprint change image is, the smaller the number of black pixels is, so that the memory amount required for the registration information can be reduced. Considering these matters, for example, one of the following specifications may be valid.

In the thinning process of the registered fingerprint image, the line width designation value is set to 1 pixel, and in the thinning process of the inspection fingerprint image, the line width designation value is set to an appropriate value of 2 pixels or more (for example, 3 pixels). The line width designation value for the thinning process of the registered fingerprint image is appropriately selected under the condition that it is smaller than the line width designation value for the thinning process of the inspection fingerprint image. In the following examples, the case will be mainly described. (7) Registration processing means of registration information of registration image The registration processing of fingerprint information is input to the image 10 of the image memory 4 as a registration fingerprint, and the registration fingerprint in the image 10 which is the result of the processing up to the thinning processing is performed. Change template Rth to sub-template RT (0) and non-sub-template RB
In this process, (0) is extracted and stored in each file. A procedure R is a procedure for performing fingerprint information registration processing. The input information of the procedure R is the file names of the registered fingerprint sub template and non-sub template, the registered fingerprint change image Rth, the fingerprint boundary information of the registered fingerprint, and the approximate center point (X _RC , Y _RC ) of the registered fingerprint. .. The output information of procedure R is
It is a file of sub-template RT (0) and a file of non-sub-template RB (0). (Procedure R) Step R1 (Creation of Sub Template RT (0)):
From the registered fingerprint change image Rth, the black pixel address in the fingerprint area and in the range of the sub template RT (0) is extracted to create a file of the sub template RT (0). The registration fingerprint approximate center point (X _RC , Y _RC ) is also stored in the storage file of RT (0). Step R2 (Creation of non-sub template RB (0)): From the registered fingerprint change image Rth, the sub template R
Black pixel addresses outside T (0) and within the fingerprint area FA are extracted to create a file of non-sub-template RB (0). (End of procedure R) The storage format of the data of each file of the sub-template and the non-sub-template is arbitrary. For example, the data may be compressed and stored in a file, and the data may be expanded at the time of use. (7) Matching Processing Means of Registered Image and Inspection Image In the matching processing, the inspection fingerprint is input to the image 10 of the image memory 4 and the processing up to the thinning processing is performed. Each with black pixels,
This is a process of checking the matching with each black pixel of the black pixel set stored in the memory 6 as the registration information regarding the registered fingerprint.

The coordinate transformation by rotating and translating the coordinate axes for aligning the registered fingerprint and the inspection fingerprint may be performed on either one of the images, but in this embodiment, the registered fingerprint is more However, the registered fingerprint is moved to match the inspection fingerprint on the assumption that the number of black pixels is reduced because the line width designation value of the thinning process is small. The outline of the matching process will be described below.

Sub Template Matching (Primary Matching) The sub template matching is performed at the position where the black pixel of the registered fingerprint change image and the black pixel of the inspection fingerprint change image are the best match in the registered fingerprint sub template RT (0). Is a process for obtaining. That is, first, for the sub-template RT (0) of the registered fingerprint change image, the sub-template RT (0, H, V) when the approximate center point of the registered fingerprint is matched with the approximate center point of the inspection fingerprint, It is calculated by moving the coordinate axis of the registered fingerprint in parallel. Next, in the vicinity of the center, when the coordinate axes of the sub-template RT (0, H, V) are rotated and moved in parallel in the vertical and horizontal directions, the inspection fingerprint change image and the registered fingerprint when the black pixels most match Of the sub-template RT (S, H, V) and the parallel movement amount (horizontal movement amount H, vertical movement amount V) (S, H,
V is an integer).

Collation of Non-Sub Template (Secondary Collation) Collation of the non-sub template is performed by S, H, V of RT (S, H, V) of the registered fingerprint obtained by the collation of the sub template.
The coordinate conversion of the black pixel address of RB (0) of the registered fingerprint is performed to obtain the black pixel address, and the match with the black pixel address of the inspection fingerprint change image is checked to determine the match between the registered fingerprint and the inspection fingerprint. This is a process of outputting information. That is, first, using the angular rotation amount S, the horizontal movement amount H, and the vertical movement amount V of the coordinate axis of the sub-template of the registered fingerprint obtained by the sub-template matching, the black color of the non-sub-template RB (0) of the registered fingerprint is black. Pixel coordinate conversion is performed to obtain RB (S, H, V). Next, the match between the black pixel of RB (S, H, V) of the registered fingerprint change image and the black pixel of the inspection fingerprint change image is checked.

The number of black pixels in the mismatched portion is checked. Final Judgment Match the registered fingerprint and the test fingerprint for all fingerprint areas based on the matching of the sub-template, the matching of the non-sub-template, and the check result of the mismatched portion.

Based on the outline of the above collation processing, procedure C shows the procedure for performing collation processing. The input information of the procedure C is the file name of the sub-template and non-sub-template of the registered fingerprint, the inspection fingerprint change image, and the approximate center point of the inspection fingerprint. The output information of the procedure C is the collation result. (Procedure C) Step C1 (Sub-template verification): (Claim 3
Step C1a : The sub template RT (0) is stored in the memory 6 from a file. Next, sub-template R
T (0), S = S _{min to} S _max , (Incremental step width K of S
), H = H _{min to} H _max , (incremental step width K of H),
And V = V _{min to} V _max , (incremental step width Kv of V), an image consistency check auxiliary procedure (procedure W) described later is executed.

As a result, S, H and V are respectively set to S _min
~ S _max , H _min ~ H _max , V _min ~ V _max ,
Change with increments Ks, Kh, Kv, suboptimal S,
The values of H, V, Sa, Ha, Va are calculated. Step C1b : The change range of S, H, and V is made smaller, and the step size is made smaller than Ks, Kh, and Kv.

S: Minimum value (Sa-Ds), maximum value (Sa
+ Ds), increment step width Ksb H: minimum value (Ha-Dh), maximum value (Ha + Dh), increment step width Khb V: minimum value (Va-Dv), maximum value (Va + Dv), increment step width Kvb Is executed to obtain {Sb, Hb, Vb} which is the optimum value of {S, H, V}. Where Ds, D
h and Dv are constants for defining the movement range (see Note C (1)). As a result, the sub-template matching rate T
The optimum values of {S, H, V} where 1 is maximum and the sub-template matching rate T1 = N1m / N1c are obtained. Next, regarding the predetermined constant TK1, if T1 ≧ TK1, it is determined that the registered fingerprint and the inspection fingerprint match by sub-template matching, and the process proceeds to step C2. If T1 <TK1, the registered fingerprint and the inspection fingerprint are It is determined that they do not match, and the procedure C is ended. (If it is moved from the beginning with a small step size, the processing amount remarkably increases. However, since the consistency is first checked and the near-optimal value obtained thereby is finely changed, the processing amount is small. ) Step C2 (collation of non-sub-template): Non-sub-template RB (0) and the optimum S, H, V obtained in step C1 are used as input information to execute the image consistency check auxiliary procedure (procedure W). To do. As a result, a non-subtemplate matching rate T2 = N2m / N2c is obtained. Next, regarding the predetermined constant TK2, if T2 ≧ TK2, it is determined that the registered fingerprint and the inspection fingerprint match, and the process proceeds to step C3. If T2 <TK2, it is determined that the registered fingerprint and the inspection fingerprint do not match. , The procedure C is completed. Step C3 (matching of mismatched portion): It is necessary to check the mismatch between the black pixel of the registered fingerprint and the black pixel of the inspection fingerprint, and exclude the case where there are too many black pixels in the mismatched portion of the inspection fingerprint change image. Therefore, the following process is performed in order to approximately obtain and determine the ratio of the black pixels in the non-matching portion when the line width of the inspection fingerprint change image is matched with the line width of the registered fingerprint change image.

Step C3a : RT (0) and RB
From the range of (0), the approximate range of the collation target area of the registered fingerprint change image after conversion of {S, H, V} is obtained. The coordinates (X ′, Y ′) in the range after conversion from the coordinates (X, Y) are
Similar to the procedure W, the following formula is used. X ′ = (X−X _RC ) · cos (S) + (Y−Y _RC ) · sin
(S) + X _TC- HY '=-(X-X _RC ) ・ sin (S) + (Y-Y _RC ) ・ co
s (S) + _YTC- V where cos (•) and sin (•) represent trigonometric functions.

Step C3b : (corresponding to the invention of claim 4) The collation target area after the coordinate conversion (that is, RT (S, H,
V) and RB (S, H, V) union) The number of black pixels Tnw of the inspection fingerprint change image is counted. That is, Tnw = total number of black pixels in the collation target area after coordinate conversion of the inspection fingerprint changed image. At this time, assuming that the line width of the inspection fingerprint changed image is w, the total number of black pixels Tnc when the line width (line width λ) of the registered fingerprint changed image is approximately Tnc = Tnw / (w / Λ). Especially when λ = 1, Tnc = Tnw / w. Nm = number of matching black pixels in the matching target area after coordinate conversion of the registered fingerprint change image and inspection fingerprint changing image Nc = total number of black pixels in the matching target area after coordinate conversion of the registered fingerprint change image is already I'm looking for it. At this time, as the degree of non-coincidence of black pixels, for example, the non-coincidence portion rate is set to Tz = (Tnc-Nm) / Nc, and in the case of 0≤Tz≤Tkc, the registered fingerprint and the inspection fingerprint are coincident as the final collation determination. If not, it is determined that they do not match. Here, Tkc, (0 ≤ Tkc ≤ 1) is a constant indicating the allowable ratio of mismatched black pixels of the inspection fingerprint image, and the smaller the value, the more severe the condition.

Conventionally, if Nm / Nc is large, it is determined that they match. However, even if Nm / Nc is large, the truth may not match. In this case, the mismatch portion rate Tz becomes large, and the mismatch is not treated as a match.
(End of Procedure C) FIG. 7 is a conceptual diagram of the relationship of parameters in step C3. FIG. 7 shows the relationship between the matching portion and the non-matching portion when the line width of the inspection fingerprint changed image and the line width of the registered fingerprint changed image are made to coincide with each other in the newly registered fingerprint area. Explaining. Remark C (1): A comparatively large range is examined by using each value of the increment step width (Ks, Kh, Kv), which is the first movement step width in step C1a, as a rough value, and the value obtained in step C1a is {S, H, V} in a relatively small range including the sub-optimal value is checked as fine values for each of the increments (Ksb, Khb, Kvb) of the second movement increments in step C1b. When the range of movement is increased, it is possible to reduce the amount of processing as compared with the case where fine increments are used for all. It should be noted that it is also possible to perform the processing by defining the range to be increased and the range to be investigated in three or more steps by repeating the processing in two steps. (8) Image Consistency Check Auxiliary Procedure The processing outline of the image consistency check auxiliary procedure (procedure W) is as follows. For each pixel address (X _R , Y _R ) of the sub-template RT (0) or non-sub-template RB (0) for the registered fingerprint, the registered fingerprint (X _R , Y
Approximate center point of _R) (X _RC, the Y _RC), test fingerprint (X _T, approximate center point of Y _T) (X _TC, balanced movement so as to coincide with the Y _TC). Next, the coordinate axis of the registered fingerprint is rotated, and the converted black pixel address (X _R $, Y _R $) is
It is checked whether black pixels are present in the fingerprint area of the inspection fingerprint change image, and parallel movement is also performed. In case of RT (S, H, V),
S, H, V and T1, N1m, N1c when the matching rate T1 at each value of S, H, V becomes maximum are obtained. RB
Regarding (S, H, V), there is only one S, H, V, and the value at that time gives the maximum T2. Since T1 and T2, N1m and N2m, and N1c and N2c are common to the primary and secondary collations in this procedure, T and Nm
, Nc.

The input information of the procedure W is the designated portion of the registered fingerprint change image (either RT (0) or RB (0)).
Black pixel address set, coordinate axis angle conversion amount S (minimum value, maximum value, increment size), horizontal movement amount H (minimum value, maximum value, increment size) of registered fingerprint coordinate axis, registered fingerprint coordinate axis Vertical movement amount V (minimum value, maximum value, increment size),
And an inspection fingerprint change image. The output information of procedure W is
Optimal coordinate axis rotation angle S of registered fingerprint, optimal coordinate axis horizontal movement amount H, optimal coordinate axis vertical movement amount V, designated area (RT (0)
Or either RB (0)) is the number Nm of matching black pixels of the registered fingerprint changed image and the inspection fingerprint changed image, the total number of black pixels Nc of the registered fingerprint changed image of the designated area, and the matching rate T.
In this procedure, up to a certain number of black pixels are checked, and if the matching rate is less than or equal to the specified value for censoring, the processing for the pattern at that time is aborted (the invention of claim 1). Further, when the matching rate is equal to or higher than the specified value for determination by the input information specification, {S, H, V} at that time is determined as the optimum value (or sub-optimal value), and the subsequent matching process is omitted. It is possible (the invention of claim 2).

The processing procedure of procedure W is shown below. FIG. 8 shows a schematic flow chart of the procedure W. (Procedure W) Step W1 (Selection of Angle S): The angle S is sequentially selected from the minimum value to the maximum value of S in increments of S according to the input information (that is, the minimum value of S is S _min , Maximum value is S
_max , when the increment is Ks, S = S _min , S _min
+ Ks, ..., S _max ), step 2
Go to Step W2 (coordinate conversion by angle S): For all black pixel addresses (X _R , Y _R ) of the input registered fingerprint thin image black pixel (either RT (0) or RB (0)) Te, and _{X R $ = X R -X RC} + X TC Y R $ = Y R -Y RC + Y TC when S = 0. When S ≠ 0, the black pixel set (RT (0)
Or, for all black pixel addresses (X _R , Y _R ) for either RB (0)), the approximate center point (X _RC , Y _RC ) of the registered fingerprint is the approximate center point of the test fingerprint. (X
After the parallel movement to align with _TC , Y _TC ), (X _TC , Y _TC ).
The coordinate axis is rotated about the angle S by the angle S. This means that X _R $ = (X _R −X _RC ) · cos (S) + (Y _R −Y _RC ).
・ Sin (S) + X _TC Y _R $ =-(X _R- X _RC ) ・ sin (S) + (Y _R-
This can be done by Y _RC ) ・ cos (S) + Y _TC . As a result, a set of all black pixel addresses (X _R $, Y _R $) of the newly registered fingerprint when H = V = 0 is obtained.

As described above, the registered fingerprint coordinate axis is rotated S degrees about the registered fingerprint approximate center point (X _RC , Y _RC ),
Further, a set of all black pixel addresses (X _R $, Y _R $) of the newly registered fingerprint when horizontal movement amount H = vertical movement amount V = 0 is obtained. (Refer to remark W (3).) Step W3 (calculation of matching rate T): Step W3a : The matching black pixel number counter Nm and the registered fingerprint total black pixel number counter Nc are initialized to 0 respectively.

Step W3b : For each address of the set of (X _R $, Y _R $), the inspection fingerprint thin image is examined, and if it is a black pixel in the fingerprint area, 1 is added to the coincident black pixel number counter Nm. , And the registered fingerprint black pixel number counter Nc
Also add 1. If it is a white pixel in the fingerprint area or outside the fingerprint area (not treated as a black pixel or a white pixel), 1 is added to the registered fingerprint black pixel number counter Nc. Here, in the sub-template matching process, it is checked whether {S, H, V} at this time can be abandoned during matching (corresponding to the invention of claim 1). That is, for a constant {Nci, Tci; i = 1, 2, ..., K}, when counter Nc = Nci (Nci and k are constants), then S,
Nm at this time is defined as Nmi, (i = 1, 2, ...) As the degree of matching up to the middle of the pattern determined by the values of H and V.
.., k), Nmi / Nci <Tci, (i = 1, 2, ..., k) (Tci and k are constants. See Note W (1).) Since it is unlikely to do so, the S, H, and V at that time are abandoned midway, and go to step W4 to go to the next value of S, H, and V. In this way, when the degree of match is poor, that is, when the degree of match is within the specified range,
Since the subsequent processing for S, H, and V is not executed, the processing amount is reduced accordingly.

Furthermore, in the matching process of the sub-template,
When there is a midway decision designation (for example, it can be configured to be designated by the input information of the procedure W), the following processing is performed (corresponding to the invention of claim 2). When the counter Nc = Ndj, Nm at this time is taken as Nmj, (j = 1, 2, ...,) As the degree of matching up to the middle of the pattern determined by the values of S, H, and V at this time. g)
For each j, Nmj / Ndj> Tdj, (j = 1, 2, ..., G) (Ndj, Tdj, and g are constants. See Note W (2)). Even if it is not, since the matching rate is sufficient, step W3c is executed and {S,
H, V} is determined as {S, H, V} of the output information.

Step W3c : For all addresses in the set (X _R $, Y _R $), when step W3b is completed, T = Nm / Nc is calculated. Then, Nm and NcT are stored for {S, H, V} at this time. Step W4 ( Translation by H and V): For the newly registered fingerprint black pixel address set (X $, Y $) when H = V = 0, H and V set in the H and V storage areas are sequentially set. When selected (when H = V = 0, already calculated in step 3) and sequentially changed from the minimum value to the maximum value of H and from the minimum value to the maximum value of V at each increment step. (That is, H has a minimum value H _min and a maximum value H _min .
_max , when the increment is Kh, H = H _min , H _min
+ Kh, ..., The maximum value is changed to H _max , and V is V _min , V _min + Vv when the minimum value of V is V _min , the maximum value is V _max , and the increment is Kv.・ ・ ・
.., (maximum V _{max is} changed by), for each {S, H, V}, (X $ -H, Y $ -V) becomes the newly registered fingerprint black pixel address set after translation,
The same processing as in step W3 is performed for each combination of {S, H, V}, and the matching rate T is stored. Step W5 (check unprocessed S): When there is an unprocessed S value, go to step W1.

When there is no unprocessed S value, step W6
go to. Step W6 (determination of maximum matching rate): each {S, H,
For V}, S, H, V and T, Nm, Nc when T = Nm / Nc is maximum are obtained and used as output information. When only one S, H, and V is input, the matching rate at that time is output information.
(End of procedure W) Remark W (1): Tci in step W3b, (i =
The values of 1, 2, ..., K) are 0 ≦ Tci ≦ 1,
For example, it is defined as follows. If the total number of black pixels of the registered fingerprint change image is Nc, then Nci / Nc, (i = 1, 2, ...
., K) represents the progress of the processing, and the possible range is 0≤Nci / Nc≤1. Nci increases and Nc
As Nmi / Nci approaches Nm / Nc, which is the coincidence rate for {S, H, V} examined at this time, Tci can be set larger as Nci is larger. The larger Tci, the wider the range of abandonment on the way and the greater the effect of reducing the amount of processing, but on the other hand, erroneous recognition tends to occur, so it is necessary to set an appropriate value.
The maximum value of the calculation is determined by the set value of k, but if the abandonment is performed once during the collation, it is not necessary to calculate the value of {S, H, V} at that time. Is. The specific value of Tci needs to be determined depending on the characteristics of the target image. Further, the conditional expression defining the range of abandonment during verification is not limited to the example shown in the procedure W as long as it determines the degree of matching up to the middle. Remark W (2): Ndj and Tdj, (j in step W3b
The values of = 1, 2, ..., G) are determined as follows, for example. The value of Ndj is Nc or less with respect to the total number of black pixels Nc of the registered fingerprint change image, and is a value close to Nc. If Nc is not known in advance, the estimated value may be used. The value of Tdj is smaller than 1 and close to 1. If the value of Tdj is reduced, the processing reduction effect is increased, but misrecognition is likely to occur. The maximum number of calculations is determined by the set value of g.
If it is performed once, the calculation of whether or not the decision can be made halfway is unnecessary. For example, {S, H,
This can be used to reduce the processing time of the first step when the increment of V} is increased in two steps.

The conditional expression for defining the range of the decision on the way of collation is
If you want to determine the degree of agreement up to the middle
The example is not limited to. Remark W (3): Step W
The expression of 2 has the following meaning. In step W2,
Less (X_R, Y_R) Approximate enrolled fingerprints for all
Center point (X_RC, Y_RC) Is the approximate center point (X
_TC, Y_TC), The new address after translation is X_R# = X_R-(X_RC-X_TC) Y_R# = Y_R-(Y_RC-Y_TC) And (X_R#, Y_R#) Becomes the new address. Next
, (X_TC, Y_TC) Around the angle S coordinate rotation
To do. This is X_R$ = (X_R# -X_TC) ・ Cos (S) + (Y_R# -Y
_TC) ・ Sin (S) + X_TC = (X_R-X_RC) ・ Cos (S) + (Y_R-Y_RC) ・ Sin
(S) + X_TC Y_R$ =-(X_R# -X_TC) ・ Sin (S) + (Y_R#-
Y_TC) ・ Cos (S) + Y _TC =-(X_R-X_RC) ・ Sin (S) + (Y_R-Y_RC) ・ Co
s (S) + Y_TC Can be obtained by Where the trigonometric function sin
The values of (•) and cos (•) are calculated in advance for the variation of the angle S.
The range value may be stored in the memory 6. (10) Flow of Registration Process and Matching Process FIG. 9 shows an outline flow of the fingerprint registration process and the matching process.
In the registration process, the registration information of the fingerprint is stored in the memory of the image inspection device 1.
This is the process of registering in 6. The verification process is the inspection fingerprint
This is a process for determining the matching of prints. From entering the fingerprint, enter
An outline of the flow up to recording or verification is shown in Procedure Z below. (Procedure Z) Steps ZA1 to ZA5 are the registration process.
This is a process that is common to processing and collation processing.Step ZA1 : Use your fingerprint from the image input device 2 to make an image memo
Fill in 4.Step ZA2 : Of the fingerprint in the image 10 of the image memory 4
Smooths a grayscale image.Step ZA3 : Binarization and background of image 10 by procedure B
Perform separation.Step ZA4 : Approximate center of fingerprint image in image 10
The point is obtained by the procedure Q.

(End of step ZA1 to step ZA4)
Subsequent processing is divided into registration processing and verification processing. Steps ZR1 and ZR2 are for the registration process,
The registration information of the registered fingerprint is registered in the memory 6. Step ZR1 : The registered fingerprint binary image (first image) in the image 10 is subjected to a thinning process in the fingerprint area to obtain a registered fingerprint changed image (first changed image). Step ZR2 : Processing of registration information of registered fingerprint (procedure R)
I do.

(End of step ZR1 to step ZR2)
Steps ZC1 to ZC2 are the case of the matching process, and the registered fingerprint and the test fingerprint are matched. Step ZC1 : The inspection fingerprint binary image (second image) in the image 10 is thinned within the fingerprint area to obtain an inspection fingerprint changed image (second changed image). Step ZC2 : By the matching process (procedure C, procedure W),
Determine the match between the registered fingerprint and the test fingerprint.

(End of step ZC1 to step ZC2)
The present invention is not limited to the embodiments described above, and can be applied to the following expansions or modifications, for example. For the image input method, smoothing process, binarization process, background separation process, correction process, approximate center point finding process, narrowing process, and matching rate and mismatch rate calculation formulas in matching processing, etc. , The claims of the present invention are
The present invention is not limited to this example, and modifications, expansions, or partial omissions using other methods (for example, known methods) are possible. When the positional deviation is so small as to be negligible, the positional alignment may be determined by the positional deviation of only a possible parallel movement and the matching rate when the matching rate is the best. The formula for obtaining X _R $ and Y _R $ in the step W2 of the procedure W can be used as long as it is a conversion that allows rotation and parallel movement, and is not limited to the present embodiment. For _{example, X R $ = X R ·} cos (S) + Y R · sin (S) Y R $ = - X R · sin (S) + Y R · cos (S) can also be used. Also, the use of coordinate transformation or geometric transformation is free. By adding the value obtained by rotating or translating the sub-template as registration information, the processing amount for collation can be reduced (in this case, the memory amount increases).

In this embodiment, the case where the image is a fingerprint has been described, but the present invention can be applied when the image can be regarded as being composed of lines. In the embodiment, the collation of one inspection fingerprint and one registered fingerprint is described, but the present embodiment is also performed when searching for the best matching registered fingerprint from two or more registered fingerprints for one inspection fingerprint. It can be applied by repeating the collation of examples.

[0082]

The present invention has the following effects. By means of abandonment during collation, useless collation can be abandoned on the way, so that the processing amount can be reduced. By means of collation decision
Since unnecessary collation is unnecessary, the processing amount can be reduced. Two
By using the step-by-step collation means to set the movement amount of the image information in two steps, the first step of moving the image information in a relatively large movement amount and collating the image information in a relatively large range without lowering the recognition accuracy. It is possible to perform the collation in the second step of collating the image information by moving the image information with a relatively small movement amount in a relatively small range. Therefore, it is possible to reduce the entire processing amount as compared with the case where one image information is moved with a small moving amount in the entire range.
The extension to three or more stages can be realized by repeating the two-stage collation, and is effective when the image is large.

By performing the check of the non-coincidence portion together with the check of the coincidence portion, there is an effect that the erroneous recognition is reduced. By separating and setting a plurality of sub-template areas, it is possible to reduce the amount of processing for alignment as compared with the case where continuous sub-template areas are used, without lowering the recognition accuracy. By separating the area of the non-subtemplate from the area of the subtemplate and setting one or more of them, it is possible to reduce the file amount for registering the non-subtemplate portion.

By limiting the range of the approximate center point, it is possible to easily align a fingerprint image with a clear center point and to collate a fingerprint image with an unclear center point. is there. The determination of the effective area has an effect that the fingerprint boundary area can be easily determined. The image inspection apparatus has an effect of easily inputting an image with good image quality when the image input is confirmed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a fingerprint identification system according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram of an image memory and use of the memory.

FIG. 3 is an explanatory diagram of a fingerprint area.

FIG. 4 is an explanatory diagram showing another example of a fingerprint area.

FIG. 5 is a diagram showing an example of an effective partial area table of a fingerprint.

FIG. 6 is an explanatory diagram showing an example of means for obtaining an approximate center point of a fingerprint image.

FIG. 7 is an explanatory diagram of checking a mismatched portion.

FIG. 8 is a flowchart showing an outline of an image consistency check assisting procedure (procedure W) and a means for abandoning during collation.

FIG. 9 is a flowchart showing an outline of fingerprint input processing, registration processing, and matching processing.

Claims (9)

[Claims] 1. A collation processing means for checking the coincidence after aligning both images, two images of a binary image having black pixels and white pixels as constituent elements of image pixels are defined as a first image and a second image. In the case where one image is moved by means of arbitrary conversion into a pattern of image information after movement and the two images are collated for one or more patterns, collation is performed for each pattern. An image inspection apparatus comprising means for calculating the degree of coincidence up to the middle of the pattern, and when the degree of coincidence is within the prescribed range of abandonment on the way, the subsequent processing for the pattern is not executed. 2. Matching processing means for checking the coincidence after aligning both images, two images of a binary image having black pixels and white pixels as constituent elements of image pixels are defined as a first image and a second image. In the case where one image is moved by means of making a pattern of image information after movement by arbitrary conversion and the two images are collated with respect to a set of one or more patterns, the individual patterns are A means for calculating the degree of coincidence halfway during the collation, and when the degree of coincidence is within the prescribed range of the intermediate determination, the pattern is regarded as the best in the set of patterns at that time, and An image inspection device characterized by not performing collation. 3. Matching processing means for checking the two images of a binary image having black pixels and white pixels as constituent elements of image pixels as a first image and a second image, and checking the coincidence after aligning the two images. When one image is moved by means of a pattern that has been moved by arbitrary conversion and the two images are collated, the pattern is moved within the first movement range with the first movement step width. By the first step and the second step in which the pattern is moved by the second moving step width within the second moving range including the sub-optimal moving position obtained in the first step, An image inspection apparatus, which is capable of obtaining various moving positions. 4. Two images of a binary image having black pixels and white pixels as constituent elements of pixels of the image are defined as a first image and a second image, and a set of one or more black pixels is treated as a line. ,
A matching processing means for checking the coincidence after the alignment of both images is provided, and one or more thinning processing means for narrowing the line width of most of the lines by designating the line width of the binary image to the line width specified value or less is provided. And the individual black pixels of the first modified image obtained by processing the line width of the first image by the thinning processing unit by setting a line width specification value less than the original line width and the second image. With respect to each black pixel of the second modified image obtained by processing the line width by setting the specified value of the line width by the thinning processing means, the pixel address in each black pixel of at least one of the modified images is set to zero times. The above-described conversion is provided with a matching processing unit for checking the matching after the alignment, and the degree of the mismatch of the black pixels in the mismatched portion is determined by comparing the black pixel mismatch when the line width of one changed image is matched with the line width of the other. As a means of approximating the degree Ri, the image inspection apparatus characterized by examining the inconsistency. 5. Matching processing means for checking the two images of a binary image having black pixels and white pixels as constituent elements of the image pixels as a first image and a second image, and checking the coincidence after aligning the two images. In a provided image inspection apparatus, a sub-template portion that is a subset of images for aligning one image area with one image area, and a non-sub-template portion that is a subset of images other than the alignment portion The image inspection apparatus is characterized in that at least one of the sub-template portion and the non-sub-template portion is divided into a plurality of pixel sets and set. 6. A collation processing means for checking the coincidence after aligning both images, two images of a binary image having black pixels and white pixels as constituent elements of image pixels are defined as a first image and a second image. The area of one image is divided into a sub-template part, which is a subset of images for aligning both images, and a non-sub-template part, which is a subset of images other than the part for which alignment is performed. The image inspection apparatus is characterized in that at least one sub-template portion and at least one non-sub-template portion are set separately by a pixel set. 7. An image inspection apparatus having means for determining an approximate center point of an image, wherein a limit range of the approximate center point is set, and when the approximate center point is out of the limit range, it is within the limit range. An image inspection apparatus characterized in that the approximate center point is set by means for setting the approximate center point within a limited range from the approximate center point. 8. A binary image is provided with a means for obtaining a brightness average value for each pixel group of an image having a brightness of two or more values, and when the image is a grayscale image, it is binarized to form a binary image. Is divided into one or more pixel sets, a brightness average value of the pixel set is obtained, and when the brightness average value is within a specified range, it is determined that the pixel set is an effective region of an image. Image inspection device. 9. An image memory is provided with a means for inputting an image having a luminance of two or more values and a means for obtaining an average luminance value for each pixel group of the image. Value image, the binary image is divided into one or more pixel sets, the brightness average value is obtained for each arbitrary pixel set of the binary image,
When the average brightness value is within the specified range, it is determined that the pixel set is an effective area, and when the number of pixel sets determined as the effective area is equal to or more than the specified number, the image in the image memory is determined. Characteristic image inspection device.