JP2004326830A - Image processing apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus for improving image processing properties. <P>SOLUTION: The image processing apparatus includes means for registering one or more addresses of holes having a size within a predetermined range during execution of registering processing for a binary registered image and a binary test image. When a hole is detected in the test image within a predetermined range from the address position of a registered hole during collating processing of the registered image and test image, the hole is judged to be concordant. It is a necessary condition for judgment of concordance with regard to the registered image and test image that, when all registered holes are examined for concordance, the number of concordant holes exceeds a predetermined ratio to the total number of registered holes. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention provides an image processing apparatus (that is, an electronic computer, an electronic exchange, a communication control apparatus, an IC card, an image processing apparatus) for analyzing and recognizing digitized images (fingerprints, imprints, figures, characters, etc.). When performed by hardware and / or software in a recognition device, an image matching device, an image inspection device, etc., one or more image processing means for improving performance (improving matching accuracy, reducing processing amount, reducing data amount) are provided. The present invention relates to an image processing device.

  As an example of an image targeted by the present invention, a case where the image is a fingerprint will be described. A fingerprint is a pattern of ridges on a finger. Since the valley line (between the ridges) is determined by the ridge, a pattern drawn by the valley may be used as a fingerprint instead of the ridge. A line treated as a fingerprint is called a fingerprint line. The fingerprint input device for personal identification includes a prism method (for example, see Non-Patent Document 1) and a hologram method (for example, non-patent method) as a method of inputting from an imaging device (for example, a CCD (Charge Coupled Device) camera). Reference 2).

  A fingerprint image of analog information input from the imaging device is converted into a digitized gray-scale image of the fingerprint by an A / D (analog / digital) converter. The grayscale image of the fingerprint is represented by coordinates (X, Y), which are the addresses of the pixels in the image memory, and the luminance of the pixels, which are components of each pixel address in the image memory. The setting method of the X axis and the Y axis is arbitrary. The fingerprint unevenness may be directly converted into a binary image to obtain a fingerprint image.

  The density image of the fingerprint can be corrected by smoothing or the direction of the ridge. The characteristic points representing the characteristics of the fingerprint include an end point, a branch point, and an intersection. The feature points of the digitized fingerprint grayscale image are obtained by binarizing the fingerprint image and further thinning the image to obtain a range of pixels representing the feature points (for example, a set of 3 × 3 pixels centered on the feature points). This can be detected by the presence of the same pattern as the pattern in the thinned image (for example, see Non-Patent Document 3).

  In fingerprint matching, a fingerprint in which information for matching is registered in a memory in advance is a registered fingerprint (registered image or registered fingerprint image), and a fingerprint to be compared with the registered fingerprint is a test fingerprint (test image or test fingerprint). Image). As a method of comparing a registered fingerprint and a test fingerprint, a method using a feature point, a method using the direction of a ridge, and a method using pattern matching between original images of a registered fingerprint and a test fingerprint are known. As a method based on pattern matching between thinned images, Patent Literature 1 describes a collation method by superimposing a thinned image of a registered fingerprint and a thinned image of a test fingerprint.

  Smoothing is a process for reducing the influence of noise in a fingerprint image. For example, there is a local weighted average filter that uses values of neighboring pixels of an arbitrary pixel, and is described in Non-Patent Document 4 and the like.

  Thinning of a binary image is to set the line width of most (mostly, from half or more to all, ideally all) lines of pixels of the type to be lined to one pixel In this case, either a black pixel or a white pixel can be selected as a target type of pixel. As a method of binarizing a gray-scale image and thinning the binary image, a black pixel on the outside in a set of black pixels is sequentially stored while maintaining the connectivity (4 or 8 connections) of the black pixels. (See, for example, Non-Patent Documents 5, 6, and 7). Non-Patent Document 8 also describes a method for thinning a grayscale image or a binary image.

  A method of binarizing a grayscale image into a binary image is described in, for example, Non-Patent Document 9. As a binarization method, a P-tile method is known, in which a ratio of a binarized target pixel (either a black pixel or a white pixel) to the total number of pixels is determined in advance. Is a method of determining a threshold value for binarization so as to be equal to the threshold value (for example, see Non-Patent Document 10). Not directly applicable.

  In the input of the fingerprint, a positional shift (rotation and translation) occurs between the test fingerprint and the registered fingerprint. Therefore, in the collation of the test fingerprint and the registered fingerprint, it is necessary to align the positions of both fingerprints. As the alignment method (rotation and translation), a method using the ridge direction, a method using representative feature points and peripheral feature points, trial and error for the range where only translation is possible, and final setting when the number of matches is the largest A method for determining the position is known. Formulas for coordinate transformation and geometric transformation necessary for alignment are described in Non-Patent Document 11, for example.

  It is useful to find an approximate center point of the fingerprint image in the alignment at the time of collation. Patent Literature 2 describes a method in which a direction in which a gradient of a ridge line is steep is sequentially searched to obtain a center point. Non-Patent Document 12 describes a method of sequentially approaching the center position by using the number of intersections of each side of a rectangle with parallel lines. In Non-Patent Document 13, the distribution of the number of lines is obtained by counting the number of ridges passing for each scanning line.

  Non-Patent Document 14 proposes a method in which a black pixel obtained from a thinned image of a registered fingerprint (or an image subjected to a thinning process) and a binary image of a test fingerprint (or an image subjected to a thinning process) are collated. In this method, the amount of processing and the amount of memory are reduced as compared with the method of comparing two binary images.

  In storing the registration information in the memory, it is necessary to reduce the storage amount as much as possible. In the present invention, it is necessary to store the thinned binary image (line figure) as registration information. Non-Patent Document 15 is known as a method of storing a line figure, but it is difficult to apply the method to a complicated case such as a fingerprint.

JP-A-63-132386, "Fingerprint collation method". Japanese Patent Publication No. 58-55548, "Method of determining center position of figure". 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). Igaki et al., "Personal collation device using holographic fingerprint sensor", IEICE Technical Report, PRU87-31, pp. 139-143. 27-33, (1987). Sasakawa et al., "Fingerprint Verification Device for Personal Confirmation with Enhanced Capability to Handle Low-Quality Images," IEICE Transactions, Vol. J72-D-ll, No. 5, pp. 707-714 (1990). Takagi and Shimoda (supervised) "Image Analysis Handbook", pp. 538-548, The University of Tokyo Press (1991). Tamura (supervised) "Introduction to Computer Image Processing", Soken Publishing, pp. 80-83 (1985). Tamura, "Study on Multi-faceted Image Processing and its Software System", Research Report of The Institute of Electronics, Technology, pp. No. 25-64, 835 (February 1984) Mori et al., "Basics of Image Recognition [1]", pp. 65-71, Ohmsha (1986). Kobayashi, "Image thinning and feature point extraction", IEICE Technical Report, PRU90-149, pp. 33-38 (1991). Mori et al., "Basics of Image Recognition [1]", PP. 37-47, Ohmsha (1986). Description on page 503 of "Image Analysis Handbook (published by The University of Tokyo, 1991)" by Takagi and Shimoda (supervised). Plastock et al., Translated by Koriyama, "Computer Graphics", pp. 84-88, McGraw-Hill Book (1987). Ito et al., "A Classification Method of Fingerprint Images Focusing on the Center Point", IEICE Technical Report, PRU89-79, pp. 139-157. 15-22 (1989). "A Study on Reference Point Extraction in Fingerprint Verification", 1987 IEICE Information and System Division National Convention, No. 125. Kobayashi, "Consideration of Fingerprint Image Matching Method," IEICE Technical Report, PRU91-45, IEICE, pp. 25-30 (July 1991). A method using Freeman's chain code (for example, "Image Information Processing" written by Yasui-in and Nakajima, pp. 113-114, Morikita Publishing (1991).

In order to improve the performance of image analysis and recognition, the amount of processing is reduced, the accuracy of recognition is improved, and the amount of registered information is reduced.
For example, conventionally, in the collation of fingerprint images, the processing with a large processing amount is restoration of a fingerprint line, thinning, and alignment of the collation. In the matching method using the feature points (end points, branch points, intersections) of the image, it is difficult to perform the matching when the amount of processing for restoring the fingerprint line is large and the feature points are unclear or the number of feature points is small. In the matching method based on pattern matching between the original image of the registered fingerprint and the original image of the inspection fingerprint, the width of the ridge of the fingerprint at the time of imprinting fluctuates due to the pressure and dryness of the finger, so that erroneous identification is likely to occur. Therefore, the storage amount for storing the registration information increases.

  In order to solve the above-mentioned problem, in the configuration according to the first aspect, the registration image and the inspection image, which are binary images, have a hole of a specified range during the execution of the registration processing unit for the registration image. Means for registering one or more addresses, and when the registered image and the inspection image match processing means are being executed, when a hole exists in the inspection image within a specified range from the address position of the registered hole, Is a match, and all the registered holes are checked.The conditions necessary for judging that the registered image and the inspection image match are that the matched hole has a specified ratio or more to the total number of registered holes. It is characterized by having means for performing.

  In the configuration according to the second aspect of the present invention, the registered image and the inspection image, which are binary images, are not included in the non-hole portion, which is a portion where a hole of a specified range does not exist, during execution of the registration processing means. Means for registering one or more addresses, and, during execution of the registered image and inspection image collation processing means, for a specified range from the registered non-hole address, when there is no non-hole in the inspection image, the non-hole It is necessary to determine whether the registered image and the inspection image match by checking all registered non-holes and determining that the matched non-holes are equal to or greater than the specified ratio with respect to the total number of registered non-holes. It is characterized by having means for setting conditions.

  Further, in the configuration according to the third aspect, the black pixel and the white pixel are the components of the pixel of the image, and each pixel address is defined as the black pixel and the black pixel in the binary image represented by the X coordinate and the Y coordinate. Any one of the pixel types of white pixels is set as a target pixel type, the position is arbitrary, a pixel set having a specified number of pixels and a shape, a type of a partial pixel set formed by a part of the pixel set, and The pixel set is composed of a representative pixel and a non-representative pixel, and the binary image is regarded as a set of pixel sets. Is checked, and only when the target pixel type exists, the X coordinate of the representative pixel, the division bit of whether the representative pixel is a black pixel or a white pixel, the black pixel and the white pixel of each non-representative pixel Bits and parts of The classification data of the pixel set is characterized in that it comprises means for storing the binary image.

  Further, in the configuration according to claim 4, for two images of a binary image in which a black pixel and a white pixel are constituent elements of the image pixels, a matching processing unit for checking the coincidence after the two images are aligned. , A moving section in at least one direction is divided into a plurality of sub-sections in a moving range in the positioning to form a partial positioning section, and the m-th (m is a positive constant) partial positioning from the beginning When the maximum value of the degree of coincidence in the section is out of the prescribed condition, the matching processing is not performed in the subsequent partial positioning section, and the matching is determined to be a mismatch.

  Further, in the configuration according to claim 5, for two binary images having a black pixel and a white pixel as constituent elements of the image, the matching processing means for checking the coincidence after alignment of the two images, In the movement range in the positioning, a section of movement in at least one direction is divided into a plurality of partial sections to form a partial positioning section, and a collation process is sequentially performed for each of the divided partial positioning sections to make a final determination. When the first image is determined to be a match, the matching is not performed in the remaining partial registration section, and the two images are provided with means for determining a match.

  Further, in the configuration according to claim 6, for two binary images having a black pixel and a white pixel as components of an image pixel, the matching processing means for checking the coincidence after aligning the two images, In the movement range in the positioning, at least one movement section in one direction is divided into a plurality of partial sections to form a partial positioning section, and one of the moving positions in each of the partial positioning sections is compared with one of the moving positions. When the set of target pixels of an image is successively separated and sequentially checked for consistency with the target pixel of the other image, each time the check of each separated target pixel set is completed, each of the separated target pixel sets is checked. The degree of matching up to the middle is determined according to the matching degree check condition of, and if they do not match, the matching at the moving position is determined to be mismatching, and the matching of the rest of the moving position is checked. It is characterized in that it comprises means to abandon.

  Further, in the configuration according to the seventh aspect, for the registered image and the inspection image that are binary images, a unit that creates registration information from an image obtained by performing a narrowing process on the registered image, Means for setting the ratio of the number of pixels to the total number of pixels within a black pixel ratio prescribed condition; means for determining whether the degree of coincidence between the registration information and the inspection image is within the coincidence prescribed condition; Means for determining whether or not the degree of image mismatch is within the degree-of-mismatch-determining condition, and determining the degree of coincidence in the collation, the degree of matching is within the degree-of-match-determining condition, and the degree of mismatch is determined by the degree of mismatch A feature is provided that means that the condition is satisfied is a condition necessary to pass in the determination of the coincidence between the registered image and the inspection image.

  Further, in the configuration according to the eighth aspect, for the inspection image and the registered image that are binary images, the registered image is compared with the registered image during execution of the matching processing unit that compares the black pixel of the inspection image with the black pixel of the registered image. Is a black pixel, the address A of the inspection image is a white pixel, and an address in a specified range near the address A of the inspection image is checked. Are determined to be approximately the same as the black pixels of the inspection image, and in the subsequent comparison of the inspection image with the registered image, the black pixels that are determined to be approximately the same as the inspection image are determined as the black pixels of the registered image. It is characterized in that it comprises means for eliminating the determination of coincidence with a black pixel other than the address A.

  Further, in the configuration according to the ninth aspect, there is provided a means for inputting and holding a plurality of images, a means for collating arbitrary two images, The two are selected sequentially in combination, one is registered as the registered image, the other is checked as the inspection image, the degree of coincidence is stored, and the registered image when the degree of coincidence is determined to be good is input. It is characterized by comprising means for selecting as a registered image from among the plurality of images that have been set.

  In the configuration according to the tenth aspect, a matching process is performed to find one registered image that matches one inspection image from an arbitrary number N (N ≧ 2) of registered images. In this case, for the registration section between the registered image and the inspection image, the original registration section is separated so as to be a union of all divided ranges, and the comparison between one inspection image and the N registered images is performed. It is characterized in that a means is provided for sequentially performing the alignment for each divided range until one registered image matching one inspection image is found.

  In the configuration according to the eleventh aspect, for two binary images, the registration information of the registered image and the image whose luminance is inverted are registered during execution of the registration processing unit for the registered image. Means for registering the registration information of the inverted registration image, and then comparing the inspection image with the registration information based on the registration image during execution of the matching processing means, and inverting the brightness of the inspection image. Means for comparing the reversal inspection image with the registration information based on the reversal registration image, and determining that both of the two comparison results are acceptable, as a condition necessary for determining that the registration image and the inspection image match. It is characterized by having.

[Action]
Image quality improvement for improving image processing performance is performed by means of a check (refer to claims 1 and 2) based on the existence position of holes or no holes by a set of white pixels in a set of black pixels. Means for correcting the image distortion (see claim 8), means for making the ratio of black pixels constant and using the degree of coincidence and the degree of non-coincidence as conditions for matching (see claim 7), The problem is solved by selectively using a means for registering the best image obtained by mutually comparing one or more images (see claim 9).

  In order to improve the performance of image processing, the amount of collation processing is reduced by collating up to m partial registration sections (m ≧ 1, m is a predetermined constant) from the beginning, and determining the maximum degree of matching between them. When the value is equal to or less than the specified value, a means for determining a mismatch (see claim 4), determination of matching of the collation is sequentially performed for each partial positioning section, and if a match is determined, the whole is determined to be a match. Means that does not perform collation for the subsequent partial alignment sections (see claim 5), the determination of matching consistency is sequentially performed for each partial alignment section, and the degree of matching is determined in the middle of the partial alignment section. If it can be determined that the value is not more than the specified value, means for performing the processing of the next partial alignment section without performing the collation for the remaining part in the partial alignment section (see claim 6), and one inspection image and Match In order to reduce the amount of collation processing when performing a process of finding one registered image from an arbitrary number of registered images, the collation with each registered image is performed by dividing the alignment section into division ranges and performing collation. The problem is solved by selectively using means (see claim 10) that are sequentially applied to the device as needed. The reduction in the amount of registered data for improving the performance of image processing can be solved by selectively using means (see claim 3) for storing type data of a partial pixel set as a pixel set shape identifier.

  If it is not necessary to calculate the degree of mismatch (unmatched portion ratio) in the collation to improve the performance of the image processing, the information of the inverted image is also used as the inversion registration information. (See Claim 11).

According to the present invention, there is an effect that the performance of image processing can be improved by selecting and applying one or more means described in the claims and the embodiments.
According to the matching means described in the first and second aspects and the embodiments thereof, since a part corresponding to a sweat gland is checked at the time of a fingerprint, there is an effect that more precise matching can be performed.
In the means relating to storage of a binary image described in claim 3 and the embodiment, when the target pixel type does not exist in the data of the partial pixel set smaller than the selected pixel set, the registered data amount can be reduced. There is an effect that can be done.

In the means relating to the matching described in claim 4 and the embodiment, at the stage where the matching from the beginning to the m-th (m = 1, 2,...) Partial matching section is performed, the matching of the entire positioning section is performed. In some cases, it can be determined that they do not match, so that there is an effect that the processing amount of collation can be reduced.
In the means relating to the collation described in claim 5 and the embodiment, the collation can be determined to be the same as the whole collation in the collation in units of the partial alignment section, so that the amount of collation processing can be reduced. There is.

According to the means relating to the matching described in claim 6 and the embodiment, in some cases, it is possible to abandon the discrepancy determination for one moving position in the partial alignment section by checking up to the middle of the black pixel set of the registration information. In addition, since this check need only be performed at the boundary of the division of the black pixel set, there is no overhead due to the check, and there is an effect that the processing amount of the collation can be reduced.
In the apparatus for executing the means described in claim 7, the ratio of the black pixels of the inspection image can be made constant with respect to the fluctuation of the input grayscale image, so that there is an effect of increasing the certainty of the collation determination.

According to the collation means described in claim 8 and the embodiment, information on the vicinity of black pixels of the registered image is also checked against the fluctuation due to the distortion of the image, so that there is an effect that the distortion of the line of the image can be dealt with. .
According to the means relating to the registration processing described in claim 9 and the embodiment, the best image can be selected as a registered image from a plurality of input images, so that the coincidence at the time of matching can be improved. There is.

In the means relating to the collation processing described in claim 10 and the embodiment, by performing the collation by dividing the moving range of the alignment into the divided ranges, by sequentially applying the collation to each registered image, This has the effect of reducing the amount of collation processing for finding one registered image that matches one inspection image from an arbitrary number of registered images.
In the case where the means described in claim 11 is selected, there is an effect that the calculation of the unmatched portion ratio can be omitted.

  As an embodiment, a case where an image is a fingerprint (sometimes called a fingerprint image) will be described. FIG. 1 is a configuration example of a fingerprint identification system. The fingerprint input from the image input device 2 is processed in the image processing device 1. The image processing device 1 includes an image memory 4 for storing digitized fingerprint grayscale images, binary images, and images subjected to various types of processing when necessary, a central processing unit 5 that is one or more CPUs, And a memory 6 for storing information such as programs, data, and files that are sets of data. When memory devices having different characteristics (for example, a semiconductor memory and a magnetic disk) coexist in the memory 6, information transfer between them is performed by hardware or software as necessary.

  The image memory 4 and the memory 6 are sections based on stored information, and may be realized by the same storage device. The image input device 2 includes an imaging 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). A pixel address in the image memory 4 for storing a fingerprint image which is a grayscale image of a digitized fingerprint is represented by (X, Y) by an X coordinate and a Y coordinate. The pixel address (X, Y) may be represented as a pixel (X, Y) or simply (X, Y). The part of the image memory 4 that stores one image is called a screen. The image memory 4 can have one or more screens. The pixel address may be simply referred to as an address.

  Each screen of the image memory 4 is composed of pixels. If the entire pixel address range is 0 ≦ X ≦ Xh and 0 ≦ Y ≦ Yh, the processing range further specified within this pixel address range is processed. Subject to. When a calculation including the pixel address and the number of pixels causes a decimal part to occur in the pixel address and the number of pixels, the processing is performed by rounding down, rounding off, or rounding up the decimal part. Pixel values are represented by luminance. Which part of the luminance becomes a ridge depends on the processing method of the image input device 2 and the processing of the image in the image processing device 1, and in any case, the characteristic of the luminance corresponding to the ridge is determined by the image processing device 1. The processing can be performed by setting in advance.

  A set of one or more pixels is called a pixel set. In identifying a fingerprint, a fingerprint image input from the image input device 2 for registration in the memory 6 of the image processing device 1 is input as a registered fingerprint (registered image or registered fingerprint image), and input from the image input device 2 for inspection. The fingerprint image to be inspected is called an inspection fingerprint (inspection image or inspection fingerprint image). In an image binarized into black pixels and white pixels, as a fingerprint line, either a black pixel or a white pixel can be selected as a target pixel type, and one of them can be a ridge line and a valley line. It is only necessary to correspond to each. In the present embodiment, a black pixel is treated as a fingerprint line.

  Thinning is a process in which most of the line width is set to one pixel. In the present embodiment, however, a line of a black pixel set is included in an image formed by black pixels of an original binary image. Reducing the width partially or entirely is referred to as a narrowing process, and an image obtained as a result of the narrowing process is referred to as a narrowing image. Therefore, thinning is one form of thinning processing in the present embodiment. Note that the line width is defined as the minimum distance (the number of pixels) that reaches the other edge through the line when a point at the edge of an arbitrary line is determined. Therefore, the line width is determined for each line edge position.

  FIG. 2A shows the state of image data stored in the image memory 4. On the screen of the image 10, a digitized image (binary image or grayscale image) obtained from the image input from the image input device 2 is stored. For example, an image at the time of input is stored on the screen of the image 11 and can be used for processing of the image 10. Depending on the selection of the processing means, the image 11 may not be necessary. In that case, the image memory 4 only needs to be the image 10. An image may be transferred from the image memory 4 to the memory 6 and image processing may be performed in the memory 6. FIG. 2B shows the situation of data and the like stored in the memory 6. The program and data 12 store a program and data for realizing the present embodiment, and the registration information 13 contains Then, the registration information of the registered fingerprint image is stored and saved in a file.

  The pixels of the image have luminance, and a value corresponding to the luminance is determined. The luminance at the pixel address (X, Y) is represented by f (X, Y). Depending on the image input method, there are a case where the grayscale image set in the image memory 4 is obtained by binarization, and a case where the binary image is directly set in the image memory 4.

  A grayscale image is an image in which a plurality of pixel brightnesses exist, and the range of variation in the brightness can be an arbitrary grayscale image. For convenience of explanation, in the present embodiment, numerical examples will be described. In the case of description, luminance is from 0 to 255, luminance 0 is the state with the lowest luminance (black), luminance 255 is the state with the highest luminance (white), and an intermediate value between the luminance of black and white. Can exist continuously.

  A binary image is represented only by black pixels and white pixels, but it is assumed that values corresponding to the respective luminances of the black pixels and white pixels are determined. Whether a black pixel is a high-luminance portion or a low-luminance portion is determined by a target image and an input method thereof, and may correspond to any case. In the present embodiment, when it is necessary to consider a luminance value (for example, description of a numerical example related to luminance), in a binary image, the luminance of a black pixel is 0 and the luminance of a white pixel is 255. The case is described as an example.

The logical origin and coordinate axis of the image stored in the image memory 4 can be determined independently of the physical pixel positions of the image memory 4. The X axis and the Y axis can be freely set, but for convenience of explanation, the X direction is horizontal from left to right (that is, the direction of increasing X), and the Y direction is top to bottom (that is, the direction of increasing Y). Vertical direction.
A set of one or more pixels is called a pixel set. FIG. 3A is an example of a pixel set (3 × 3 pixel set) including 3 × 3 pixels (that is, 3 × 3 pixels), and FIG. 3B is a pixel set (4 × 4 pixels) (4 × 4 pixels). FIG. 3C is an example of a pixel set of 4 × 3 pixels (4 × 3 pixel set).
In FIG. 3A, a pixel {P1, P3, P5, P7} is called a 4-neighbor pixel of P0, and a pixel {P1, P2, P3, P4, P5, P6, P7, P8} is an 8-neighbor pixel of P0. Call.

In the embodiment of the present invention, division is performed only when the denominator is not 0. The definitions of the main terms and notations are shown below. However, some of these notations may not appear in the embodiments.
[M]: represents an arbitrary number m rounded down to the decimal point.
||: Represents the concatenation of two arbitrary numbers as they are. For example, 1 || 0 || 1 = 101.
~: For any two numbers (or symbols), a to b have the meaning from a to b, and a and b are both included in the range.
f (X, Y): luminance at a pixel address (X, Y).

FA: Fingerprint area. The fingerprint boundary is treated as within the fingerprint area.
(XC, YC): The address of the approximate center point of the fingerprint.
(XRC, YRC): Address of approximate center point of registered fingerprint.
(XTC, YTC): The address of the approximate center point of the test fingerprint.
Rth: a registered fingerprint changed image obtained by performing at least a narrowing process on a registered fingerprint binary image (first image) obtained from a registered fingerprint image (shade image or binary image).

Tth: a test fingerprint changed image obtained by subjecting a test fingerprint binary image (second image) obtained from a test fingerprint image (shade image or binary image) to at least adjustment or thinning of the ratio of black pixels.
RA: A set of addresses (X, Y) of all black pixels in the fingerprint area in the registered fingerprint changed image Rth. RA is the union of RT (0), RB (0), and black pixels in the unused fingerprint area.

RT (0): RT (0) is a subset of RA. RT (0) can designate an area of one or more arbitrary number of 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 a sub-template part. The sub-template is used for image alignment (see FIG. 4).

RB (0): RB (0) is a subset of RA and can specify an area of one or more arbitrary number of separated pixel sets. RB (0) is called a non-sub template. The part of the image from which the non-sub-template is extracted may be referred to as a non-sub-template part. Non-sub-templates are not used for image registration. RT (0) and RB (0) are set so as not to overlap (see FIG. 4).

RT (S): A set of addresses (X, Y) of black pixels of the sub template RT (0) when the coordinate axis of the registered fingerprint changed image is rotated S degrees around an arbitrarily determined point.
RT (S, H, V): after rotating the coordinate axis of the registered fingerprint changed image by S degrees around an arbitrary point (for example, an approximate center point of the registered fingerprint), and after performing horizontal movement H and vertical movement V Is a set of addresses of the sub-template RT (0) on the coordinate axes of. RT (0,0,0) = RT (0) and RT (S, 0,0) = RT (S). (Note that a set of values of S, H, and V that specify the movement position determines one pattern in which the registered fingerprint changed image has been moved by coordinate transformation.)

RB (S, H, V): after rotating the coordinate axis of the registered fingerprint changed image by S degrees around an arbitrary point (for example, an approximate center point of the registered fingerprint), and after performing horizontal movement H and vertical movement V Is a set of addresses of the non-sub-template RB (0) on the coordinate axes of. RB (0,0,0) = RB (0) and RB (S, 0,0) = RB (S).
Nlm: The number of black pixels matching the sub-template, and represents the number of black pixels matched between the black pixels of the sub-template of the registered fingerprint changed image and the black pixels of the test fingerprint changed image.

Nlc: 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 changed image.
N2m: the number of black pixels matching the non-sub-template, and represents the number of black pixels that match the black pixels of the non-sub-template of the registered fingerprint changed image and the black pixels of the test fingerprint changed image.
N2c: Non-sub-template total black pixel number, which indicates the total black pixel number of the non-sub-template of the registered fingerprint changed image.
Nm: represents Nlm or N2m. The counter Nm is a counter value for calculating Nm.
Nc: represents Nlc or N2c. The counter Nc is a counter value for calculating Nc.

Partial area: An individual area when an image is divided into a plurality of parts.
Section [a, b]: Represents a section including both ends from a to b for arbitrary values a and b.
Positioning section: represents a range (a set of moving sections for each direction) in which images are moved for positioning two images, and is composed of {rotating moving section, horizontal moving section, vertical moving section}.

Partial positioning section: An individual positioning section when the positioning section is divided into one or more subsets. The entire registration section is a union of one or more partial registration sections.
Movement position: A position represented by one set of {rotational movement position, horizontal movement position, vertical movement position}.
Registered fingerprint (registered image): Registered fingerprint image or registered fingerprint change image.
Inspection fingerprint (inspection image): Inspection fingerprint image or inspection fingerprint change image.
Threshold: a constant value for making some judgment by comparing the evaluation value with the threshold. How to determine the threshold value differs for each evaluation value.

  FIG. 4 shows an example of a relationship between a fingerprint area defined by a fingerprint boundary and a part for extracting the sub-template RT (0) and a part for extracting the non-sub-template RB (0) in the image 10 in the image memory 4. . Examples 1 to 3 are examples of a sub-template portion and a non-sub-template portion.

  The means for processing a fingerprint image will be described below. If there is no description of a process to be performed next in a step in the procedure of each unit, the process proceeds to the immediately following step. As for the constants used in each means, appropriate values are statically or dynamically set in consideration of various conditions in the image processing apparatus 1 and the image input apparatus 2. The realization of each step and the processing in which they are linked can be modified as long as the contents of the processing are the same.

(1) Setting means for input confirmation trigger from the image input device to the image memory The image 10 input from the image input device 2 to the image memory 4 is a target object of the imaging device 7 (that is, if the target is a fingerprint, the image Since it changes with the movement of the finger (input to the input device 2), it is necessary to determine an opportunity to determine the image 10 in the image memory 4. Processing methods for this include:
(A) a method of notifying the image processing apparatus 1 of an image decision trigger by a signal from the image input apparatus 2;
(B) a method in which a user instructs the image processing apparatus 1 to determine an opportunity to fix an image, or
(C) a method in which the image processing apparatus 1 determines an image decision trigger by examining the state of the image 10 in the image memory 4;
There is. Here, an example of the method (c) will be described.

In a state in which the gray-scale image of the fingerprint is correctly input, the average luminance value is within a certain range, and a stable image can be obtained when this state continues. By taking advantage of this, in order to improve the quality of the image for which the input is determined, a luminance average value is determined for a predetermined pixel set of the image (for example, the entire region of the image or one or more partial regions), When the average luminance value is within the specified condition, the image is determined to be a valid image, and when this is continuously confirmed a specified number of times or more, the image of the temporarily fixed image memory 4 is finally determined. Determine.
An example of the setting means of the image input final opportunity based on the above is shown in a procedure IN.

(Procedure 1N)
Step IN1:
The counter Fa is initialized to 0.
Step IN2:
The image input process is started, and the image in the image memory is temporarily fixed.
Step IN3:
A specified area average brightness value Eav is obtained for the specified area of the image. If the specified area average brightness value is within the specified conditions, it is determined that the image is a valid image, and the process goes to step IN4. It is determined that the image is an image, and the procedure goes to step IN1. here,
The defined area average brightness value Eav = (sum of the brightness of the pixels in the defined area) / (the total number of pixels in the defined area)
The specified condition of Eav is EL ≦ Eav ≦ EH.
EL and EH are constants, and are determined in consideration of a fluctuation range of luminance when a target image is input.

When there are a plurality of specified regions, it is checked whether or not the average value of the specified regions is within the specified conditions for each specified region. It is determined that there is.
Step IN4:
Add 1 to Fa.
Step IN5:
It is checked whether Fa is the specified value Fat. If it is the specified value, the process proceeds to step IN6. If it is less than the specified value, go to step IN2. The value of the constant Fat is set in consideration of the shortest number of times until the input image is stabilized.

Step IN6:
Determine the image in the image memory. (End of procedure IN)
FIG. 5 is an outline of a flowchart of a procedure IN for continuously confirming input and validity of an image and determining an image.

(2) Image Conversion Means For an image input from the image input device 2 to the image processing device 1, the smaller the number of pixels to be processed, the better the performance of the image can be improved as long as the recognition accuracy is not reduced.
For example, when the number of pixels in the image memory is larger than the number of pixels to be processed (as a numerical example, the number of pixels in the image memory 4 is 640 × 480 pixels (X = 0 to 639, Y = 0 to 479)) If the number of pixels to be processed is 256 × 240 pixels (X = 0 to 255, Y = 0 to 239), the image is input to the image memory 4 and then input. The range of the image is examined, a subset of the image is extracted as a temporary image, and a small image corresponding to the required number of pixels to be processed is extracted from the temporary image, and the small image is newly added to the normal processing target. Image. An example of the image conversion means for this is shown in Procedure ET.

(Image conversion procedure ET)
Step ET1 (extraction of the small image in the X direction):
Steps ETla to ETld are executed.
Step ETla (determines the X-coordinate at the left end of the provisional image where black pixels exist):
Let Ka be the increase width in the X direction (for example, Ka = 1).
The following processing is performed in the order of u = 0, Ka, 2Ka, 3Ka,... until the left end of the temporary image is found.

The total value GA (u) of the luminance of the Y coordinate (Y = 0, Ja, 2Ja,...) Extracted in the Y direction when X = u was obtained, and the existence of the luminance equivalent to the black pixel could be confirmed first. Let u be the X coordinate of the left end of the temporary image. For example, if no image is input and all the images have high luminance, Ja is set to an increasing amount in the Y direction (for example, Ja = 1), and it is confirmed that GA (u) <GA. When it is completed, u at this time is set as the X coordinate of the left end of the temporary image.
(Here, in order to increase the certainty, a condition that GA (u + Ka) <GA may be further added.) The constant GA is a threshold value indicating that there is a black pixel when binarized. The luminance is determined in consideration of the luminance so that the total value of the luminance of the pixels extracted in the Y direction is close to the minimum value that can be regarded as including the black pixel.

Step ET1b (determines the X coordinate of the right end of the temporary image where the black pixel exists):
The following processing is performed in the order of v = Xh, Xh-Ka, Xh-2Ka, Xh-3Ka,... until the right end of the temporary image is found.

The total value GA (v) of the luminance of the Y coordinate (Y = 0, Ja, 2Ja,...) Extracted in the Y direction when X = v was obtained, and the existence of the luminance equivalent to the black pixel could be confirmed first. Let v be the X coordinate of the right end of the temporary image. For example, Ja is set to an increase width in the Y direction (for example, Ja = 1), and when it is confirmed that GA (v) <GA, the value of v at that time is set as the X coordinate of the right end of the temporary image.
(Here, in order to increase the certainty, a condition that GA (v−Ka) <GA may be further added.)

Step ETlc (determine the Wx of the X coordinate of the middle point with respect to the left and right ends of the temporary image):
The Wx of the X coordinate of the middle point with respect to the left end and the right end of the temporary image is determined by the following equation.
Wx = (u + v) / 2

Step ET1d (determines the start X coordinate for extracting a small image)
The starting X coordinate Xa of the small image is
Xa = Wx-XL / 2
Determined by XL is the number of pixels in the X direction of the small image. Here, in order to make the size of the small image area constant, when Xa <0, Xa = 0, and when Xa + XL-1> Xh, Xa = Xh-XL.

Step ET2 (extraction of range of small image in Y direction):
Here, in steps ET2a to ET2d, the same processing as in step ET1 is performed in the Y direction to obtain the Y coordinate range of the small image.
Step ET2a (determines the Y coordinate of the upper end of the provisional image where the black pixel exists):
Let Kb be the increase width in the Y direction (for example, Kb = 1).
The following processing is performed in the order of u = 0, Kb, 2Kb, 3Kb,... until the upper end of the temporary image is found.

The total value GB (u) of the luminance of the X coordinate (Y = 0, Jb, 2Jb,...) Extracted in the X direction when Y = u was obtained, and the existence of the luminance equivalent to the black pixel could be confirmed first. Let u be the X coordinate of the left end of the temporary image. For example, Jb is set to an increase width in the Y direction (for example, Ja = 1), and when it is confirmed that GB (u) <GB, u at this time is set as the Y coordinate of the upper end of the temporary image.
(Here, in order to increase certainty, it may be further added on condition that GB (u + Kb) <GB.) Here, GB is a threshold value for the presence of a black pixel, The luminance is determined in consideration of the luminance so that the total luminance value of the pixels extracted in the X direction is close to the minimum value that can be regarded as including a black pixel.

Step ET2b (determines the Y coordinate of the lower end of the provisional image where the black pixel exists):
The following processing proceeds in the order of v = Yh, Yh-Kb, Yh-2Kb, Yh-3Kb,... until the lower end of the temporary image is found.

The total value GB (v) of the luminance of the X coordinate (X = 0, Jb, 2Jb,...) Extracted in the X direction when Y = v was obtained, and the existence of the luminance equivalent to the black pixel was first confirmed. Let u be the X coordinate of the lower end of the temporary image. For example, Jb is set to increase in the Y direction (for example, Ja = 1), and when it is confirmed that GB (v) <GB, the value of v at that time is set as the Y coordinate of the lower end of the temporary image.
(Here, in order to increase the certainty, a condition that GB (v−Kb) <GB may be further added.)

Step ET2c (determines the Wy of the Y coordinate of the middle point with respect to the upper and lower ends of the temporary image):
The Wy of the Y coordinate of the middle point with respect to the upper and lower ends of the temporary image is determined by the following equation.
Wy = (u + v) / 2
Step ET2d (determining a start Y coordinate for extracting a small image):
The starting Y coordinate Yb of the small image is
Yb = Wy−YL / 2
Determined by YL is the number of pixels in the Y direction of the small image. Here, in order to make the size of the small image area constant, when Yb <0, Yb = 0, and when Yb + YL-1> Yh, Yb = Yh-YL.

Step ET3 (extract processing target pixels):
The range from X = Xa + XL-1 and from Y = Yb to Y = Yb + YL-1 is set as a small image to be processed.
Step ET4 (a small image is made a normal image):
Conversion for treating the range of the small image as a normal image is performed as needed.
For example, the setting of the origin, the conversion of the setting location of the image, and the like. (End of step ET)

  FIG. 6 is an example of an explanatory diagram of conversion of the processing range of an image, and shows an example of an image set in an image memory, a temporary image, and a small image. In the procedure ET, only one of the X direction and the Y direction may be executed, and the other may be set to a fixed range. When the image in the image memory is used as the image to be processed as it is, the conversion by the procedure ET is unnecessary.

(3) Smoothing means Smoothing is processing for reducing the influence of noise on a fingerprint image. A known method can be used for the smoothing process. For example, a local weighted average filter using the value of a pixel in the vicinity of each pixel can be used.
If a binary image can be directly input to the image memory 4, smoothing can be omitted. When the binary image is smoothed, the image becomes a grayscale image, so it is necessary to perform the binarization again.

(4) Binarization unit and background separation unit Binarization is a process of converting a grayscale image into a binary image. The background separation is a process for 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 is described in Procedure B. The input information in the procedure B is input image information. The output information of the procedure B is a binary image of the output image and fingerprint boundary information. When an image input device of a type that can directly input a binary image to the image memory 4 is used, the binarization (step B1) is omitted and only background separation is performed. Conversely, when only binarization is required and background separation is not required, only step B1 may be 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 may be different from the length of the partial region in the X direction, but the case where the length is equal will be described below for simplicity.
L = (Xh + 1) / K
And Here, K is selected such that L is an integer. Generally, the length of the partial area can be made variable for each partial area. next,

Kmax = ((Xh + 1) / K) -1
And For the image 10, a partial area address for uniquely identifying the partial area is set as a head pixel address of each partial area. The arbitrary pixel address (X, Y) is M = [X / K] and N = [Y / K] in the partial area address J (M, N). Therefore, the pixel address (X, Y) corresponding to the partial area address J (M, N) is
X = KM, (M = 0, 1, 2,..., Kmax)
Y = K · N, (N = 0, 1, 2,..., Kmax)
Is determined by

Step B1 (binarization):
Step B1a:
M = 0, 1, 2,..., Kmax
N = 0, 1, 2,..., Kmax
, The following processing is performed. For each partial area J (M, N), for each individual partial area, the average luminance value Bav (M, N) of all pixels in the partial area is calculated as follows:
Bav (M, N) = (sum of luminance of pixels in partial area) / (number of pixels in partial area)
Ask by here,
X = KM (M = 0 to Kmax)
Y = K · N (N = 0 to Kmax)
In (M, N) covers the range of the image 10.

That is, the range of the image is
0 ≦ X ≦ Xh
0 ≦ Y ≦ Yh
When
M = 0 to [Xh / K]
N = 0 to [Yh / K]
Target range. At this time,
The total number of partial regions of the image 10 = ([Xh / K] +1) · ([Yh / K] +1). Also,
Average luminance value of entire image = (sum of luminance of entire image) / (number of pixels of entire image).

Step Blb:
In order to reduce discontinuities caused by sub-region boundaries, conversion between sub-regions is performed. That is,
M = 1 to ([Xh / K] -1)
N = 1 to ([Yh / K] -1)
about,
Cav (M, N) = (A0 · Bav (M, N) + A1 · Bav (M + 1, N) + A2 · Bav (M + 1, N−1) + A3 · Bav (M, N−1) + A4 · Bav (M− (1, N-1) + A5.Bav (M-1, N) + A6.Bav (M-1, N + 1) + A7.Bav (M, N + 1) + A8.Bav (M + 1, N + 1)) / (A0 + A1 + A2 + A3 + A4 + A5 + A6 + A7 + A8).

A0, Al, A2, A3, A4, A5, A6, A7, and A8 are constants and are determined in consideration of the mutual relationship between partial regions. For example,
A0 = Al = A2 = A3 = A4 = A5 = A6 = A7 = A8 = 1.
Cav (M, N) for each (M, N) is stored in the work area of the memory.
next,
M = 1 to ([Xh / K] -1)
For N = 1 to ([Yh / K] -1),
By setting Bav (M, N) = Cav (M, N), it is possible to change the average luminance value Bav (M, N) of each partial region due to the mutual relationship between the partial regions (M, N). it can.

  FIG. 7 is an example of the relationship between the partial area (M, N) and the average luminance value Bav (M, N) of the partial area. Here, the method of changing the average brightness value of the partial areas based on the mutual relationship between the partial areas is not limited to the above method. For example, it is possible to increase or decrease the number of partial regions used as a correlation, or to modify the above-described expression of Cav (M, N).

Step B1c:
(A) For each partial area, a threshold value T for binarization is set for the luminance f (X, Y) of the pixel (X, Y) of the partial area for which Bav (X, Y) has been obtained.
T: partial area luminance average value + D + Da
Here, D is a constant for correcting the partial area luminance average value (for example, D = 0). Da is a variable (the initial value is, for example, Da = 0) and will be described later.

(X, Y) for the partial area (M, N) at this time, ie,
KM <X <K <(M + 1) -1
K · N ≦ Y ≦ K · (N + 1) -1
The luminance f (X, Y) for (X, Y) in the range
When f (X, Y) ≧ T, (X, Y) is set to a white pixel (however, when white and black are inverted, a black pixel is set).
When f (X, Y) <T, (X, Y) is set to a black pixel (however, when white and black are inverted, white pixels are set).

  The above is the binarization means using the average value of the partial area luminance as the threshold element. In a partial area where a large number of high-luminance pixels exist in a partial area, and in a partial area where a change in luminance is small, the luminance average value becomes a high value. May become black pixels. As a countermeasure, a partial region having a high average luminance value and a small change in luminance is binarized not by the above-described binarization threshold value but by a predetermined threshold. The specific procedure for this is shown below.

For each partial area of the original image (shade image),
Bav (・, ・) ≦ Bm
In the case of, binarization is performed using the above-mentioned average luminance value as a threshold value.
Bav (・, ・)> Bm
(Bm is a value for determining whether or not the luminance average value is high. For example, when luminance 180 to 255 is regarded as high luminance, Bm = 180). For each pixel in the region,
G = | (average luminance value of partial area) − (luminance of each pixel in partial area) |
The maximum value of G is determined as Gmax, and as a result, when Gmax ≦ Bsmall (Bsmall is the upper limit of the luminance change width. For example, the luminance change from 0 to 9 is very small. Is assumed to be Bsma11 = 10), the pixels in the partial area can be regarded as having little change in luminance, so that the partial area has a specified binarization threshold T = Bg (for example, Bg is the entire image). , Or a fixed value).
This procedure is repeated for all partial areas of the original image. FIG. 8 is an example of a schematic flowchart of a binarization process in a partial region in which the luminance slightly changes.

As the next process, the luminance state of the image differs depending on the situation at the time of input of the image. Therefore, the ratio of the black pixels in the partial area to all the pixels is obtained. When the ratio is outside the specified range, the value of Da is changed. The binarization process is repeated a specified number of times (one or more times). The procedure for optimizing the binarization threshold is as follows. The total effective partial area black pixel ratio Rbw is
Rbw = (number of black pixels in all partial areas) / (total number of pixels in all partial areas)
And

Rbw1 ≦ Rbw ≦ Rbw2
In the case of, since the black pixel ratio is normal, the binarization can be completed.
Rbw> Rbw2
In the case of, T is reduced to reduce the number of black pixels. That is, Da is replaced with (Da-Da1).
Rbw <Rbw1
In the case of, T is increased to increase the number of black pixels. That is, Da is replaced with (Da + Da1).

  Rbw1 and Rbw2 are constants for stabilizing the black pixel ratio. For example, when the black pixel ratio is desired to be a value between 40% and 45%, Rbw1 = 0.40 and Rbw2 = 0.45. I do. Here, Da1 is a constant for increasing or decreasing the threshold value for binarization (for example, Da1 = 1). In the binarization, the ratio of black pixels is obtained. When the ratio is out of the specified range, the process of changing the value of the threshold value T for binarization and binarizing is repeated a specified number of times. If the ratio of black pixels is out of the specified range even after the specified number of repetitions, for example, the processing is terminated and the processing is performed from the beginning of image input.

  Here, a plurality of different values of Da1 can be used to shorten the processing time for the black pixel ratio to fall within the specified range and to increase the probability that the black pixel ratio falls within the specified range. FIG. 9 is an example of a process for stabilizing the black pixel ratio of the binarization of the binary image divided into the partial areas. As a constant group for changing Da, Da1, Da2, Da3, (For example, Da1 = 3, Da2 = 2, Da3 = 1). Dai, (i = 1, 2,...) Can be composed of an arbitrary number of constants, and by appropriately setting the number and value of this constant group, the value of Rbw can converge to a specified range. And convergence speed can be improved. Note that Alp, A2p, A3p, Alm, A2m, and A3m in FIG. 9 are parameters that are dynamically changed during processing for selection of processing.

Step B2 (display of effective partial area):
An average partial area luminance value is obtained again for the binary image obtained as a result of step B1. next,
Partial area address IMN = ([X / K], [Y / K])
(The partial area address is the start address of each partial area.)
X = KM, (M = 0, 1, 2,..., Kmax)
Y = K · N, (N = 0, 1, 2,..., Kmax)
For each of the partial areas, it is determined whether the partial area is valid or invalid based on the average luminance value Bav (X, Y) obtained by the partial area average luminance processing means.

That is, BL ≦ Bav (X, Y) ≦ BH
, It is determined that the partial area is an effective partial area, and an effective display is set in the effective partial area table Y (M, N).
Y (M, N) = ｛1: When the area is an effective partial area. 0: In the case of an invalid partial area. ｝
Here, BL and BH are constants for distinguishing an effective partial area from an invalid partial area.

Step B3 (inspection of the number of valid partial areas):
According to Y (M, N), YT = the number of effective partial areas is counted, and if YT ≧ YC (YC is, for example, the total number of partial areas × threshold ratio constant value), the procedure goes to step B4. . If YT <YC, it is considered that the number of effective partial areas is insufficient, and this procedure ends abnormally.

Step B4 (left end of fingerprint boundary in partial area units):
The partial area fingerprint boundary information is represented by {(NT, ML, MR), NT = 0 to Kmax}. For each value of N = NT, ML ≦ M ≦ MR is a fingerprint area in partial area units.
In this step, the left end ML of the fingerprint boundary is obtained for each partial area. Starting from N = 0, the following processing is performed for N = NT, (NT = 0 to Kmax). In the increasing direction of M from the left end (M = 0), 1 (effective partial area display) in the element G (M, N) of the effective partial area table G is sequentially searched, and 1 is Kc or more (Kc is a partial The value ML of the first M of a continuous portion for determining the fingerprint boundary of the region (one or more constants) is defined as the left end of the fingerprint region for N at that time. If Mc = 0 to Kmax and 1 (effective area display) cannot be continuously found for Kc or more, the value of N is the non-fingerprint area for all M, and ML = MR for the non-fingerprint area NT. = -1.

Step B5 (right end of fingerprint boundary for each partial area)
In this step, the right end MR of the fingerprint boundary is obtained for each partial area. Starting from N = 0, for N = NT and (NT = 0 to Kmax), N for which the left end is not set in step B4 (that is, N when ML = −1) is skipped and the right end (M = Kmax), in the decreasing direction of M, sequentially search for 1 (effective partial area display) in the element G (M, N) of the effective partial area table G, and 1 is Kc or more (Kc is a constant) continuous part Is the right end of the fingerprint area for N at that time. As described above, the partial area fingerprint boundary information {(NT, ML, MR), NT = 0 to Kmax} is obtained.

Step B6 (fingerprint boundary information for each pixel):
From the partial area fingerprint boundary information {(NT, ML, MR), NT = 0 to Kmax}, fingerprint boundary information for each pixel is obtained. Fingerprint boundary information for each pixel is represented by {(YT, XL, XR), YT = 0 to Yh}, and for each YT value, XL ≦ X ≦ XR means the fingerprint area.

According to the partial area fingerprint boundary information, when N = 0 to Kmax,
For Y = YT of K · N ≦ Y ≦ K · N + Kmax,
XL = KML
XR = K ・ MR
To obtain fingerprint boundary information {(YT, XL, XR), YT = 0 to Yh} for each pixel. (End of procedure B)

  FIG. 10 is an example of an effective partial area table of a fingerprint. Hereinafter, when simply referred to as fingerprint boundary information, it means fingerprint boundary information {(YT, XL, XR), YT = 0 to Yh} for each pixel. As for the background separation, as an alternative to the steps B2 to B6, a means for simply treating the entire image as a fingerprint area is also possible.

(5) Means for Calculating Approximate Center Point In the approximate center point of the fingerprint image, a point near the center ([Xh / 2], [Yh / 2]) of the image 10 or a point near the center (XC, YC) is determined. ).
Further, any other known means for obtaining an approximate center point may be used.

(6) Thinning means Thinning is a process for reducing the line width of most of an image to one pixel. Any thinning method (for example, a known thinning method as described in the related art) may be used as the thinning means.

(7) Thinning processing means The thinning processing is to reduce the line width of most of the image to a specified line width or less. The narrowing process can be approximately realized by making the ratio of black pixels to all pixels (or the ratio of white pixels to all pixels, or the ratio of black pixels to white pixels) constant in the binarization stage. Is also regarded as one of the narrowing processing means.

Further, there are the following methods for reducing the line width of the binary image.
A set of one or more black pixels is treated as an image line. The holding method of the line width designation value is arbitrary, and can be used as input information of the thinning processing means or can be held in the thinning processing means. One or more line width designation values may be held in one narrowing processing unit. In the present embodiment, the thinning processing means can be selectively used according to the specified line width value. An example of the thinning processing means depending on the line width designation value will be described below.

(A) Narrowing processing to make most line widths one pixel (line width specified value is one pixel)
This is a means for reducing the line width of a binary image (or a means for binarizing a grayscale image and reducing the line width), and a known method (for example, a thinning method capable of setting most of the line width to one pixel). ) Can be used. There is also a method of directly binarizing and thinning a grayscale image.

(B) Thinning processing for narrowing most of the lines below the specified line width (the specified line width is an arbitrary number)
It can be realized based on a known method. For example, there are the following means.
i) The processing for one screen, in which line elements are deleted one pixel at a time from the outside of a line (corresponding to a black pixel which is a fingerprint line in the present embodiment) constituting an image, is reduced to a line width of most of one pixel. In the case of a thinning method that repeats until the thinning process,
{(Maximum line width of original binary image)-(Line width specified value)} / (Approximate number of pixels deleted for each line width in processing for one screen)
The number of repetitions may be determined so that

ii) In the case of the thinning method in which a black pixel is left from the center of a line element constituting an image: For a black pixel on a line segment where an arbitrary straight line and a line of the image intersect, a line width specification including the center of the line segment is specified. Pixels less than or equal to the value are left (if the line width is equal to or greater than the line width specified value, black pixels corresponding to the line width specified value are left around the midpoint of the line segment, and line segments whose line width is less than the line width specified value) All the upper black pixels are left).

  In the present embodiment, the line width specification value when performing the thinning process can be arbitrarily specified for the registered fingerprint image and the inspection fingerprint image, but the line quality and characteristics of the input image, the line width specification value for the thinning process And the performance required for the image processing apparatus 1. Increasing the difference between the line width designation values in the thinning processing of the registered fingerprint image and the test fingerprint image increases the displacement between the two images, but if the difference is too large, the collation accuracy may decrease. is there. The smaller the line width of the thinned image of the registered fingerprint, the smaller the number of black pixels, so that the amount of memory required for the registered information can be reduced. Considering these, for example, one of the following designations may be effective.

(A) In the thinning processing of the registered fingerprint image, the line width designation value is set to one pixel, and in the thinning processing of the inspection fingerprint image, an appropriate value of two pixels or more is included, including the case where the line width designation value remains unchanged. I do.
(B) An appropriate selection is made under the condition that the line width designation value for the thinning process of the registered fingerprint image is smaller than the line width designation value for the thinning process of the inspection fingerprint image.
(C) In the thinning process of the registered fingerprint image, the line width designation value is set to one pixel, and the line width of the inspection fingerprint image is adjusted by the ratio of the black pixels at the time of binarization. (The following example mainly describes this case.)

(8) Registration processing means for registration information of a registration image The registration processing of the fingerprint information is performed by inputting the registration fingerprint into the image 10 of the image memory 4 and registering the registration fingerprint in the image 10 which is the result of the processing up to the thinning processing. This is a process of extracting the sub-template RT (0) and the non-sub-template RB (0) from the changed image Rth and storing them in respective files. The procedure for registering fingerprint information is shown in Procedure R. The input information of the procedure R is the file names of the sub-template and non-sub-template of the registered fingerprint, the registered fingerprint changed image Rth, the fingerprint boundary information of the registered fingerprint, and the approximate center point (XRC, YRC) of the registered fingerprint. The output information of the procedure R is a file of the sub-template RT (0) and a file of the non-sub-template RB (0).

(Procedure R)
Step R1 (creation of sub-template RT (0)):
From the registered fingerprint change image Rth, a black pixel address in the fingerprint area and within the range of the sub-template RT (0) is extracted, and a file of the sub-template RT (0) is created. The stored file of RT (0) also stores the registered fingerprint approximate center point (XRC, YRC).

Step R2 (creation of non-sub-template RB (0)):
A black pixel address outside the sub-template RT (0) and within the fingerprint area FA is extracted from the registered fingerprint changed image Rth, and a file of the non-sub-template RB (0) is created.
The storage format of the data of each file of the sub-template and the non-sub-template is arbitrary. For example, data may be compressed and stored in a file, and data may be expanded when used.

Step R3:
Step R3a (creating file RSa of sweat gland hall):
If the user selects to check the sweat gland hole, this step is executed. In the registered fingerprint, for a predetermined area, an address having a hole due to a sweat gland having a certain size or more (a hole formed by a set of white pixels in a binary image in which ridges are black pixels) is determined by a hole search unit. The specified number is obtained, and the center address is recorded in the sweat gland hole file.

Step R3b (creation of file RSb of non-sweat gland hole):
If the user selects to check the non-sweat gland hole, this step is executed. The non-hole search means obtains a prescribed number of addresses having no sweat glands of a predetermined size or more in a predetermined area in the registered fingerprint, and records the center address in a non-sweat gland hole file. (End of procedure R)

(9) Hall search means (procedure WS)
FIG. 11A shows an example of the hole. For example, a hole can be detected by the following procedure WH.
(Procedure WH)
Step WH1:
The candidate addresses for searching for holes in the image are sequentially examined, and an arbitrary white pixel address A is selected. As the candidate address, for example, it is possible to select all pixels, every n pixels (n = 2, 3, 4, etc.).

Step WH2:
It is checked whether or not a specified number of connected white pixel sets AA exist in the peripheral pixels connected to the white pixel at the address A. If not, the procedure WH is executed again from the beginning. go to.
Step WH3:
It is checked whether or not the periphery of the white pixel set AA is surrounded by the connected black pixels. If not, the procedure WH is executed again from the beginning. And
Step WH4:
This procedure WH is repeated for all candidate addresses to be searched. (End of procedure WH)

(10) Non-hole search means (procedure NWH)
An example of a non-hole is shown in FIG. A non-hole can be detected, for example, by the following procedure NWH.
(Procedure NWH)
Step NWH1:
The candidate addresses for searching for non-holes in the image are sequentially examined, and an arbitrary black pixel address B is selected. As the candidate address, for example, it is possible to select all pixels, every n pixels (n = 2, 3, 4, etc.).

Step NWH2:
It is checked whether or not a specified number of connected black pixel sets BB exist in peripheral pixels connected to the black pixel at the address B. If not, the procedure NWH is re-executed from the beginning. The pixel address B is a non-hole address.
Step NWH3:
This procedure NWH is repeated for all candidate addresses to be searched.
(End of procedure NWH)

(11) Image Data Storage Means When the address (X, Y) of each black pixel is stored in a file as it is, the required storage storage amount is as follows.
Storage storage amount = number of black pixels / (unit storage amount of X coordinate + unit storage amount of Y coordinate)
The registered fingerprint image needs to be stored in a file for each of the sub-template and the non-sub-template. An example of means for compressing the amount of data and storing it in a file without greatly increasing the processing amount in order to reduce the amount of stored data as compared with the case where the black pixel address of the binary image is stored as it is will be described below.

  A case where the pixel set is 4 × 4 pixels will be described. A set of 4 × 4 pixels is shown in FIG. 3B for an arbitrary representative pixel P0. P0 to P15 indicate the division of black pixels and white pixels of each pixel by bits, and Q = P15 || P14 || P13 || ... || P7 || P6 || P5 || P4 || P3 | By | P2 || Pl || P0 (|| represents connection), the state of black and white pixels of each peripheral pixel can be displayed by a 2-byte pixel set code Q (0000 to FFFF in hexadecimal). Next, a pixel address (X, Y) based on a relative position with respect to the representative pixel P0 = (X0, Y0) is shown below with reference to FIG.

P1: X = X0-1, Y = Y0
P2: X = X0-2, Y = Y0
P3: X = X0-3, Y = Y0
P4: X = X0, Y = Y0-1
P5: X = X0-1, Y = Y0-1
P6: X = X0-2, Y = Y0-1
P7: X = X0-3, Y = Y0-1
P8: X = X0, Y = Y0-2
P9: X = X0-1, Y = Y0-2
P10: X = X0-2, Y = Y0-2
P11: X = X0-3, Y = Y0-2
P12: X = X0, Y = Y0-3
P13: X = X0-1, Y = Y0-3
P14: X = X0-2, Y = Y0-3
P15: X = X0-3, Y = Y0-3

  The compression process of the registered fingerprint image data is a process of converting (X = 0 to Xh, Y = 0 to Yh), which is the (X, Y) coordinates of the registered fingerprint image data, into the following format. For every four Y-coordinates in the fingerprint effective area, every fourth X-coordinate is used as a representative pixel, and it is checked whether or not there is a black pixel in the range of a 4 × 4 pixel set. Only when this is done, the representative pixel X coordinate and the pixel set code are stored.

  Since the X address portion is specified every four pixels, the lower two bits can be used as a pixel set shape identifier. Utilizing this, the pixel set code, when the half (lower half, upper half, or left half) is all white pixels, respectively, the other half (upper half, lower half, right half, corresponding to the order described above) ) Is expressed as a partial pixel set.

(A) When the pixel set shape identifier is “11” in bit representation The pixel set code is represented by 2 bytes.
(That is, P15 || P14 ||... | Pl || P0).
(B) When the pixel set shape identifier is “10” in bit representation The pixel set code is represented by the upper half 1 byte.
(That is, P15 || P14 || P13 || P12 || P10 || P9 || P8).

(C) When the pixel set shape identifier is “01” in bit representation The pixel set code is represented by the lower half 1 byte.
(That is, P7 || P6 || P5 || P4 || P3 || P2 || P1 || P0).
(D) When the pixel set shape identifier is “00” in bit representation The pixel set code is represented by one byte in the right half.
(That is, P13 || P12 || P9 || P8 || P5 || P4 || P1 || P0).

Here, when the pixel set code is 1 byte, all of the omitted pixels are white pixels. In the case where the pixel set code is 1 byte, the X address portion including 2 bits for the pixel set shape identifier can be set to a true X address by setting the bit portion for the pixel set shape identifier flag to 0. it can.
That is, at the time of compression, the (X, Y) coordinates (that is, X = 0 to Xh, Y = 0 to Yh) of the thinned image data are converted into the following format. For every fourth Y coordinate, every fourth X coordinate is used as a representative pixel, and it is checked whether or not a black pixel exists in the range of the 4 × 4 pixel set. The pixel X coordinate and the pixel set code are stored. At this time, an example of means for storing the binary image is shown in Procedure G.

(Procedure G)
Step G1:
A set of 4 × 4 pixels is sequentially selected by the X address portion.
It is checked whether all the pixels (P15 to P0) are all white pixels. When all the pixels are white pixels, this 4 × 4 pixel set is skipped, and the process proceeds to the next 4 × 4 pixel set, and Step G1 is executed from the beginning. If all pixels are not white pixels, the process goes to Step G2.
Step G2:
It is checked whether all the upper halves (P15 to P8) are white pixels. If the upper half is not all white pixels, go to step G3. When the upper half is all white pixels, the lower half (P7 to P0) is represented by a 1-byte pixel set code. The lower 2 bits of the X address portion are set to a bit pattern “01”.

Step G3:
It is checked whether the lower half (P7 to P0) is all white pixels. If all the lower halves (P7 to P0) are not white pixels, go to step G4. When the lower half (P7 to P0) is all white pixels, the upper half (P15 to P8) is represented by a 1-byte pixel set code. The lower two bits of the X address portion are set to the bit pattern “10”.
Step G4:
It is checked whether all the left halves (P15, P14, P11, P10, P7, P6, P3, P2) are white pixels. If the left half is not all white pixels, go to step G5. When the left half is all white pixels, the right half (P13 || P12 || P9 || P8 || P5 || P4 || P1 || P0) is represented by a 1-byte pixel set code. The lower two bits of the X address portion are set to the bit pattern “00”.

Step G5:
The 4 × 4 pixel set is represented by a 2-byte pixel set code. The lower two bits of the X address portion are left as is ("11" of the pit pattern).
Step G6:
The processing for the one pixel set code and the X address ends.
Step G7:
Steps G1 to G6 are repeated for all 4 × 4 pixel sets for each X address.

Step G8:
The file storage format of the binary image is
Y coordinate = Ys, number of sets of stored X address parts,
{(X address portion of representative pixel As (including 2 bits of pixel set shape identifier), pixel set code),
(X address portion of representative pixel Bs (including 2 bits of pixel set shape identifier), pixel set code),
…｝
…

Y coordinate = 3 + 4j, number of sets of storage x address part,
{(X address portion of representative pixel Aj (including 2 bits of pixel set shape identifier), pixel set code),
(X address portion of representative pixel Bj (including 2 bits of pixel set shape identifier), pixel set code),
…｝
…

Y = Ye = stored number of sets of X address portions ｛(x address portion of representative pixel Ah (including 2 bits of pixel set shape identifier), pixel set code),
(X address portion of representative pixel Bh (including 2 bits of pixel set shape identifier), pixel set code),
…｝, {End display symbol}
As a binary image. (End of procedure G)

  Here, the number of sets in the storage X address portion indicates the number of pixel set codes stored corresponding to the Y coordinate at that time. Packing is performed without setting the Y coordinate or the like where the number of sets of the storage X address portion is 0. A 4 × 4 pixel set of all white pixels is not recorded. The pixel set code may be one byte or two bytes, and is distinguished by the lower two bits of the X address portion. Instead of the number of sets of the stored X address portion, an end symbol of the set of (representative pixel X address portion, pixel set code) may be added for each Y coordinate value.

FIG. 12A shows an example in which the image memory is divided into small areas of a 4 × 4 pixel set. FIG. 12B is an example of the upper half of a 4 × 4 pixel set. FIG. 12C is an example of the lower half of a 4 × 4 pixel set. FIG. 12D shows an example of the right half of a 4 × 4 pixel set. Although the size of the small area when dividing the image memory into small areas is described in the case of a 4 × 4 pixel set, it can be set arbitrarily, and the division method in the small area at that time is also set arbitrarily. May be.
Further, in order to return the format of the binary image data compressed in the procedure G to the original (X, Y) format, a process reverse to the procedure G may be performed.

(12) Matching Processing Means of Registered Image and Inspection Image The matching process is performed on each black pixel set of the black pixel set of the test fingerprint changed image and the black pixel set stored in the memory 6 as registration information on the registered fingerprint changed image. This is a process of checking the coincidence with each black pixel.
The coordinate conversion by rotation and translation of the coordinate axes for aligning the registered fingerprint changed image and the test fingerprint changed image may be performed on either one of the images, but in the present embodiment, the registered fingerprint changed image is more preferable. Assuming that the number of black pixels is reduced because the line width designation value of the thinning process is small, the registered fingerprint changed image is moved to match the test fingerprint changed image. The outline of the matching process will be described below.

(A) Matching of Sub-Template Matching of the sub-template is a process of finding a position where the black pixel of the registered fingerprint changed image and the black pixel of the test fingerprint changed image best match with respect to the registered fingerprint sub-template RT (0). is there. That is, first, with respect to the sub-template RT (0) of the registered fingerprint changed 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 test fingerprint is: It is obtained by parallel movement of the coordinate axis of the registered fingerprint change image. Next, in the vicinity of the center, when the coordinate axis of the sub-template RT (0, H, V) is rotated and translated up, down, left, and right, the registered fingerprint when the test fingerprint changed image and the black pixel most coincide. The conversion angle S and the amount of parallel movement (the amount of horizontal movement H, the amount of vertical movement V) of the sub-template RT (S, H, V) of the changed image are obtained (S, H, and V are integers).

(B) Collation of non-sub-template and collation of template The collation of non-sub-template is registered by S, H, V of RT (S, H, V) of the registered fingerprint change image obtained by collation of the sub-template. The coordinate conversion of the black pixel address of the RB (0) of the fingerprint changed image is performed to determine the black pixel address, the consistency with the black pixel address of the test fingerprint changed image is checked, and the registered fingerprint changed image and the test fingerprint changed image are checked. This is a process of outputting information on the matching. That is, first, the non-sub-template RB (0) of the registered fingerprint is used by using the angular rotation amount S, the horizontal movement amount H, and the vertical movement amount V of the coordinate axes of the sub-template of the registered fingerprint changed image obtained by the collation of the sub-template. ) Is performed to obtain RB (S, H, V). Next, the matching between the black pixels of the RB (S, H, V) of the registered fingerprint changed image and the black pixels of the test fingerprint changed image is checked. From this result, the matching rate of the entire template between the registered fingerprint changed image and the test fingerprint changed image is obtained.

(C) Check the number of black pixels in the mismatched portion.
(D) If the user has selected to perform a check using sweat glands, the check is performed.
(E) Based on the above results, the final match between the registered fingerprint and the test fingerprint is determined.
The procedure for performing the collation processing based on the outline of the collation processing described above is shown in Procedure C. The input information in the procedure C is the sub-template and non-sub-template of the registered fingerprint change image, the test fingerprint change image, and the approximate center point of the test fingerprint. The output information of the procedure C is a collation result.

(Procedure C)
Step C0:
It is assumed that the maximum total range of movement of the registered image information at the time of comparison is a section [Smin, Smax] in the rotation angle direction, a section [Hmin, Hmax] in the horizontal direction, and [Vmin, Vmax] in the vertical direction. At least one section in the moving direction is divided into one or more subsections,
Smin = {Smin (Is): Is = 1, 2,..., Js}
Smax = {Smax (Is): Is = 1, 2,..., Js}
Hmin = {Hmin (Ih): Ih = 1, 2,..., Jh}
Hmax = {Hmax (Ih): Ih = 1, 2,..., Jh}
Vmin = {Vmin (Iv): Iv = 1, 2,..., Jv}
Vmax = {Vmax (Iv): Iv = 1, 2,..., Jv}
And

here,
Section [Smin, Smax], Section [Smin (1), Smax (1)], Section [Smin (2), Smax (2)], ..., and Section [Smin (Js), Smax (Js)] Is the union of
Section [Hmin, Hmax], section [Hmin (1), Hmax (1)], section [Hmin (2), Hmax (2)], ..., and section [Hmin (Jh), Hmax (Jh)] Is the union of
Section [Vmin, Vmax], Section [Vmin (1), Vmax (1)], Section [Vmin (2), Vmax (2)], ..., and Section [Vmin (Jv), Vmax (Jv)] Is the union of

Thereafter, from step C1,
{Section [Smin (1), Smax (1)], section [Smin (2), Smax (2)], ..., section [Smin (Js), Smax (Js)]},
{Section [Hmin (1), Hmax (1)], section [Hmin (2), Hmax (2)], ..., section [Hmin (Jh), Hmax (Jh)]}, and
All sections {Vmin (1), Vmax (1)], section [Vmin (2), Vmax (2)],..., Section [Vmin (Jv), Vmax (Jv)]} were selected. The partial registration sections are sequentially executed, and the matching processing is terminated when it can be determined that there is a match in an arbitrary registration section, or when it is possible to discard the subsequent matching and determine that there is no matching.

Step C1 (sub template matching):
Steps Cla to Cld are executed.
Step Cla:
The sub template RT (0) is stored in the memory 6 from the file. Next, the sub-templates RT (0), S = Smin (Is) to Smax (Is), (S increment increment Ks), H = Hmin (Ih) to Hmax (Ih), (H increment increment Kh ), And V = Vmin (Iv) to Vmax (Iv), (the increment Kv of V), and the registration information black pixel search increment increment Kr = Kra, (Kra ≧ 1), and the image coincidence described later. The check assist procedure (procedure W) is executed.
As a result, the increments Ks, Kh, and V are set for S, H, and V for Smin (Js) to Smax (Js), Hmin (Jh) to Hmax (Jh), and Vmin (Jv) to Vmax (Jv), respectively. The value is changed by Kv, and Sa, Ha, and Va, which are sub-optimal values of S, H, and V, are obtained.

Step C1b:
For S, H, and V, as respective moving ranges,
S: (Sa−Dsb) to (Sa + Dsb), increasing step size Ksb
H: (Ha−Dhb) to (Ha + Dhb), increment increment Khb
V: (Va−Dvb) to (Va + Dvb), increment increment Kvb
, The procedure W is executed with the registration information black pixel search increment step Kr = Krb, (Krb ≧ 1), and {Sb, Hb, Vb} which is the suboptimal value of {S, H, V} is obtained. Ask.
Here, Dsb, Dhb, and Dvb are constants for determining the movement range (see note C (1)).

Step Clc:
S, H, V respectively
S: (Sb−Dsc) to (Sb + Dsc), increasing step size Ksb
H: (Hb−Dhc) to (Hb + Dhc), increment Khb
V: (Vb−Dvc) to (Vb + Dvc), increment increment Kvb
, The procedure W is executed with the registration information black pixel search increment step Kr = Krc, (Krc ≧ 1), and {Sc, Hc, Vc} which is a sub-optimal value of {S, H, V} is obtained. Ask.
Here, Dsc, Dhc, and Dvc are constants for determining the moving range.

Step C1d:
S, H, V
S = Sc, Dsd = 0, increment step Ksd = 0
H = Hc, Dhd = 0, increment increment Khd = 0
V = Vc, Dvd = 0, increment Kvd = 0
As a result, the procedure W is executed with the registration information black pixel search increment step Kr = Krd, (Krd = 1), and the optimal {S, H, V} values are obtained. Here, Dsd, Dhd, and Dvd are constants for determining the moving range. As a result, optimal values of {S, H, V} that maximize the sub-template coincidence rate T1, and the sub-template coincidence rate T1 = Nlm / Nlc are obtained.

Next, for a predetermined constant Tk1,
T1 ≧ Tk1
If so, the registered fingerprint and the test fingerprint are determined to match by matching the sub-template, and the process goes to step C2,
T1 <Tk1
If so, it is determined that the registered fingerprint and the test fingerprint do not match in the currently executing partial registration section, and the processing of the next partial registration section is executed from the beginning of step C1. When there is no partial registration section to be executed, it is determined that the registered fingerprint and the test fingerprint do not match, and the procedure C is terminated.

When Tm1 is the maximum value of T1 in the first to m-th partial positioning sections, and when Tm1 <Tmk1 (j), the subsequent processing of this partial positioning section and the remaining partial positioning The matching process for the section is not executed because it is unlikely to be a match even if it is executed, and the registered fingerprint and the test fingerprint are determined to be mismatched, and the procedure C ends.
Here, Tmk1 (j), (j = 1, 2,..., M) are predetermined constants so that the above determination can be made.

Step C2 (collation of non-sub-template and collation of template):
The image matching check assisting procedure (procedure W) is executed using the non-sub-template RB (0) and the optimal {S, H, V} obtained in step C1 as input information. As a result, N2m and N2c are obtained, and the template matching rate T2 = (Nlm + N2m) / (Nlc + N2c) is obtained using the result of step C1.

T2 ≧ Tk2
If so, it is determined that the registered fingerprint and the test fingerprint match, and the process goes to step C3.
T2 <Tk2
If so, it is determined that the registered fingerprint and the test fingerprint do not match in the partial registration section being executed, and the processing of the next section is executed from the beginning. When there is no partial registration section to be executed, it is determined that the registered fingerprint and the test fingerprint do not match, and the procedure C is terminated.

Step C3 (collation of mismatched part):
It is necessary to examine the mismatch between the black pixel of the registered fingerprint and the black pixel of the test fingerprint to eliminate the case where the number of black pixels in the mismatched portion of the test fingerprint changed image is too large. Therefore, in order to approximately determine and determine the ratio of the black pixels of the mismatched portion when the line width of the test fingerprint binary image is matched with the line width of the registered fingerprint changed image, the processing of steps C3a to C3b is performed. Do.

Step C3a:
From the range of RT (0) and RB (0), an approximate range of the matching target area of the registered fingerprint changed image after {S, H, V} conversion is obtained. The coordinates (X ′, Y ′) of the range after the conversion from the coordinates (X, Y) are calculated by the following equation, similarly to the procedure W.
X ′ = (X−XRC) · cos (S) + (Y−YRC) · sin (S) + XTC−H
Y ′ = − (X−XRC) · sin (S) + (Y−YRC) · cos (S) + YTC−V
Here, cos (•) and sin (•) represent trigonometric functions.

Step C3b:
The number of black pixels Tnw of the test fingerprint changed image in the matching target area after coordinate transformation (that is, the union of RT (S, H, V) and RB (S, H, V)) is counted. That is,
Tnw = the total number of black pixels in the verification target area after the coordinate conversion of the test 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 this is set to the line width (line width λ) of the registered fingerprint changed image is approximately:
Tnc = Tnw / (w / λ)
It is.

Here, when the line width of the registered fingerprint changed image is one pixel due to thinning, the test fingerprint changed image may also be thinned to obtain the number of black pixels Tnw of the verification target area of the test fingerprint changed image, In that case, W = λ = 1. Also,
Nlm + N2m = the number of matching black pixels in the matching target area after coordinate conversion of the registered fingerprint changed image and the test fingerprint changed image Nlc + N2c = the total number of black pixels in the matching target area after coordinate conversion of the registered fingerprint changed image has already been obtained I have.

At this time, as the degree of mismatch of black pixels, for example,
Tz = (Tnc−Nlm−N2m) / (Nlc + N2c)
age,
| Tz | ≦ Tkc
In the case of, the registered fingerprint and the test fingerprint are determined to be acceptable with respect to the unmatched portion ratio, and otherwise, are determined to be mismatched.

This is a constant indicating the allowable ratio of mismatched black pixels. (Since Tz ≧ 0 is usually used, different values of Tkc are used for Tz ≧ 0 and Tz <0, respectively, so that Tz <0 is more strict than Tz ≧ 0. May be.)
If they do not match, it is determined that the registered fingerprint and the test fingerprint do not match in the currently executing partial registration section, and the processing of the next partial registration section is executed from the beginning of step C1. When there is no partial registration section to be executed, it is determined that the registered fingerprint and the test fingerprint do not match, and the procedure C is terminated.

Step C4 (partial matching of sweat glands):
In order to improve the matching accuracy, at least one of Step C4a and Step C4b of Step C4 can be selectively added.
Step C4a (check of holes by sweat glands):
For addresses obtained by converting the addresses of the holes by the optimum {S, H, V}, sweat glands (one or more in the image in which the ridge portion of the fingerprint becomes a black pixel by binarization) Check if there is a hole composed of white pixels). The size of the white pixel set to be checked at this time is a parameter constant. next,
T4a = (number of matching hole addresses) / (total number of registered hole addresses)
In search of
T4a ≧ Tk4a
If, the sweat gland hole check passes.

Step C4 (check for non-hole):
For the addresses obtained by converting the non-hole addresses with the optimum {S, H, V}, it is checked that there are no sweat glands at those addresses in the test fingerprint changed image. The size of the white pixel set to be checked at this time is a parameter constant. next,
T4b = (number of non-hole address matches) / (total number of registered non-hole addresses)
In search of
T4b ≧ Tk4b
, The non-hole check passes. Here, Tk4b is a constant of the threshold value.

Step C5 (final judgment):
If all of the selected and executed sub-template matching rates, template matching rates, non-matching rate, and sweat gland checks are passed for the executed partial alignment section, the registered fingerprint and the test fingerprint are determined to match, and the procedure is performed. Terminate C. If not, the registered fingerprint and the test fingerprint are determined to be mismatched in the currently executing partial registration section, and the processing of the next partial registration section is executed from the beginning of step C1. When there is no partial registration section to be executed, it is determined that the registered fingerprint and the test fingerprint do not match, and the procedure C is terminated. (End of procedure C)

FIG. 13 is an example of a schematic flowchart of the collation processing based on the procedure C. FIG. 13 illustrates an example in which only the section in the rotation angle direction S is divided to form a partial positioning section.
FIG. 14 is an example of an explanatory diagram for checking a mismatched portion in the matching process, and illustrates a relationship between a matched portion and a mismatched portion when the line width of the test fingerprint changed image matches the line width of the registered fingerprint changed image. ing.

  Remarks C (1): A relatively large range is examined by using the values of the increment increments (Ks, Kh, Kv), which are the first movement increments in step Cla, as coarse values, and the ΔS, obtained in step Cla is obtained. In a relatively small range including the sub-optimal values of H and V}, the values of the increment increments (Ksb, Khb, Kvb), which are the second movement increments in step Clb, are examined as fine values, thereby making it possible to perform positioning. When the range of movement is increased, the processing amount can be reduced as compared with the case where a fine increment is used for all of the movement ranges.

  In step C1 of procedure C (three-stage multi-stage), the increment of each value of movement can be made larger than 1 in the first and second stages, and is determined in the previous stage in the second and third stages. Based on the sub-optimal values of S, H, and V, the matching may be performed within a range determined by the line width or the like in the preceding stage. In the first and second stages, the black pixels of the registration information are searched for. By setting the step width Kr to 2 or more, the number of black pixels of the registered fingerprint can be limited by skipping by a certain number, so that the processing amount of collation (substantially proportional to the number of alignment search) can be reduced. The maximum value and the minimum value (Smin, Smax, Hmin, Hmax, Vmin, Vmax) of each moving range are determined based on the maximum allowable moving range of the finger at the time of fingerprint input. Step Cld is for determining T1 in step C1.

Here, there are the following properties. The number of searches increases as the moving range increases. The smaller the increment size, the greater the number of searches. The increment step width in the middle stage is determined in consideration of the increment step width immediately before and immediately after each step. Except for the last stage, jump search is possible. Inappropriate settings of the moving range, the increment, and the jump search, erroneous recognition is likely to occur.
In step C3b, the line width of the test fingerprint changed image was approximated to the line width of the registered fingerprint changed image to determine the unmatched portion ratio, but the line width of the ascending fingerprint changed image was approximated to the line width of the test fingerprint changed image. Then, it is also possible to obtain the unmatched portion rate.

(13) Image Matching Check Auxiliary Procedure The process outline of the image matching check auxiliary procedure (procedure W) is as follows. For each pixel address (XR, YR) of the sub-template RT (0) or the non-sub-template RB (0) for the registered fingerprint, the approximate center point (XRC, YRC) of the registered fingerprint (XR, YR) is determined. The translation is performed so as to coincide with the approximate center point (XTC, YTC) of (XT, YT). Next, the coordinate axis of the registered fingerprint is rotated, and it is checked whether the converted black pixel address (XR #, YR #) is a black pixel in the fingerprint area of the test fingerprint changed image, and is also translated.

  In the case of RT (S, H, V), S, H, V, and T1, Nlm, Nlc at which the matching rate T1 at each value of S, H, V is maximized are obtained. For RB (S, H, V), there is only one S, H, V. Note that T1 and T2, Nlm and N2m, and Nlc and N2c are referred to as T, Nm, and Nc, respectively, because they can be handled in substantially the same manner in the procedure W.

  The input information of the procedure W includes a set of black pixel addresses of a designated portion (either RT (0) or RB (0)) of the registered fingerprint changed image, an angle conversion amount S of the coordinate axis (minimum value, maximum value, increment in increments). Width), the horizontal movement amount H (minimum value, maximum value, increment) of the coordinate axis of the registered fingerprint change image, the vertical movement amount V (minimum value, maximum value, increase increment) of the coordinate axis of the registered fingerprint, the inspection fingerprint change The number of jumps J in the image and the registered fingerprint black pixel skip search. Here, the movement range is an entire section or a partial section.

The registration information black pixel search increment increment Kr specifies an increment for searching for a registered fingerprint changed image black pixel when matching a registered fingerprint changed image black pixel with a test fingerprint changed black pixel. For example, Kr = 1 When Kr = 2, all the registered fingerprint changed image black pixels are searched for, and when Kr = 2, every other registered fingerprint changed image black pixel is searched.
The output information of the procedure W is, for the input information, one of the optimal coordinate axis rotation angle S, the optimal coordinate axis horizontal movement amount H, the optimal coordinate axis vertical movement amount V, and the designated area (RT (0) or RB (0)) of the registered fingerprint. ), The number of matching black pixels Nm of the registered fingerprint changed image and the test fingerprint changed image, the total number of black pixels Nc of the registered fingerprint changed image in the designated area, and the matching rate T.

FIG. 15 is an example of a schematic flowchart of an image consistency check assisting procedure (procedure W) in the matching processing. An example of the procedure of the procedure W is shown below.
(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 in the section specified by the input information, and the procedure goes to step W2. (That is, when the minimum value of S is Smin (Is), the maximum value is Smax (Is), and the increment is Ks, up to S = Smin (Is), Smin (Is) + Ks,..., Smax (Is) Change it.)

Step W2 (coordinate conversion by angle S):
With respect to the black pixel set (either RT (0) or RB (0)) of the input registered fingerprint changed image, the target black pixel address (XR, YR) to be searched by the registered information black pixel search increment Kr. ),
(A) When S = 0 XR ＠ = XR−XRC + XTC
YR ＠ = YR−YRC + YTC
And

(B) When S ≠ 0 For all black pixel addresses (XR, YR) for the black pixel set (either RT (0) or RB (0)) of the input registered fingerprint changed image, After the parallel movement for adjusting the approximate center point (XRC, YRC) of the registered fingerprint to the approximate center point (XTC, YTC) of the test fingerprint, the coordinate axis rotation of the angle S about (XTC, YTC) is performed. This means
XR ＠ = (XR−XRC) · cos (S) + (YR−YRC) · sin (S) + XTC
YR ＠ = − (XR−XRC) · sin (S) + (YR−YRC) · cos (S) + YTC
Can be performed. Thereby, a set of all black pixel addresses (XR ＠, YR ＠) of the newly registered fingerprint when H = V = 0 is obtained.

  As described above, when the coordinate axis of the registered fingerprint is rotated by S degrees around the approximate center (XRC, YRC) of the registered fingerprint, and all the blacks of the new registered fingerprint when the horizontal movement amount H = the vertical movement amount V = 0 are set, A set of pixel addresses (XR #, YR #) is obtained.

Step W3 (calculation of coincidence rate T):
Step W3a:
The coincident black pixel number counter Nm and the registered fingerprint changed image total black pixel number counter Nc are each initialized to zero.
Step W3b:
For each address of the set of (XR ＠, YR ＠), check the test fingerprint change image,
(A) If the pixel is a black pixel in the fingerprint area, 1 is added to the coincident black pixel number counter Nm, and 1 is also added to the registered fingerprint black pixel number counter Nc.
(B) If it is a white pixel in the fingerprint area or outside the fingerprint area (not a black pixel or white pixel), add 1 to the registered fingerprint black pixel number counter Nc.
Here, in the collation processing of the sub-template, it is checked whether or not the collation of {S, H, V} at this time can be abandoned.

That is, since the registered fingerprint information to be examined is a set of black pixel addresses, it is divided into k continuous sections in the search order, and every time a section i (i = 1, 2,..., K) ends,
Counter Nc = Nci
Assuming that Nm at this time is Nmi, (i = 1, 2,..., K), the degree of matching up to the middle of the pattern determined by the values of S, H, and V is Nmi / Nci. From the constant {Tci; i = 1, 2,..., K},
Nmi / Nci <Tci, (i = 1, 2,..., K)
(Tci and k are constants; see Remarks W (1).) Since subsequent checks are not expected, S, H, and V at that time are discarded halfway and the next S, H , V to go to step W4.

The next process (search for neighboring pixels at the time of matching) can be selectively performed. (This neighborhood pixel search processing can be selectively applied to each stage of the sub-template RT and to the non-sub-template RB.)
When the black pixel of the registered fingerprint change image at an arbitrary address (A) in the image is a black pixel at the address A of the test fingerprint change image, it is called a perfect match.

  When the black pixel of the registered fingerprint change image at an arbitrary image address (referred to as A) is a white pixel at the same address A of the test fingerprint change image, an address near the address A of the test fingerprint change image is checked. If there is, the black pixel of the registered fingerprint changed image at the address A is determined to be the same as the black pixel of the test fingerprint changed image (when it is necessary to distinguish it from a perfect match, it is called a neighborhood match). At this time, when the black pixel of the registered fingerprint changed image is collated, the black pixel of the same address in the test fingerprint changed image is not referred to twice or more, and the same address of the test fingerprint changed image is completely checked. The matching black pixel is not referred to more than once. For this purpose, the following operation is performed.

  The test fingerprint changed image is saved in the work area of the memory. The black pixel of the test fingerprint changed image determined to be coincident with the black pixel of the registered fingerprint changed image by perfect match or neighborhood match is determined by setting the pixel value of the image memory to an intermediate value other than the luminance value of the black pixel and the white pixel (for example, When the brightness of the black pixel is 0 and the brightness of the white pixel is 255, the brightness of the perfect match is Ba (a value other than the brightness of the black pixel and the white pixel, for example, 50), and the brightness of the neighborhood match is Bb (for example, 50). , 100)). In the comparison of the black pixel of the registered fingerprint changed image with the black pixel of the test fingerprint changed image, the black memory of the test fingerprint having the same address in the same image is not referred to twice. Check that When the comparison of the registered fingerprint change image and the test fingerprint change image by one set of {S, H, V} is completed, the test fingerprint change image is returned to the original state based on the save information of the work area.

Step W3c:
When step W3b is completed for all addresses of the set of (XR ＠, YR ＠) and the input is the sub-template RT (0),
T = Nm / Nc
Is calculated. Then, Nm, Nc, and T are stored for {S, H, V} at this time.

Step W4 (translation by H and V):
With respect to the newly registered fingerprint black pixel address set (X ＠, Y と き) when H = V = 0, H and V set in the H and V storage areas are sequentially selected (when H = V = 0, Has already been calculated in step W3) and sequentially changed from the minimum value of H to the maximum value and from the minimum value of V to the maximum value at each increment (that is, the minimum value of H is Hmin, the maximum value of H When the value is Hmax and the increment is Kh, the maximum value is changed to Hmax (Ih) by H = Hmin (Ih), Hmin (Ih) + Kh, etc. The minimum value of V is Vmin (Iv) and the maximum value Is Vmax (Iv) and the increment is Kv, V = Vmin (Iv), Vmin (Iv) + Kv,..., Is changed to the maximum Vmax), and for each {S, H, V}, X ＠ -H, Y ＠ -V) Since the recording fingerprint black pixel address aggregation, the combination of the individual {S, H, V}, performs the same processing as step W3.

Step W5 (check for unprocessed S):
When there is an unprocessed value of S, the process goes to step W1.
If there is no unprocessed value of S, the procedure goes to step W6.
Step W6 (judgment of maximum matching rate):
In the case of the sub-template RT (0), T = Nm / Nc becomes the maximum for each value of S and each {S, H, V} due to the change of H = Hmin to Hmax and V = Vmin to Vmax. S, H, V, and T, Nm, Nc at that time are obtained and used as output information. Even when only one S, H, and V is input, T, Nm, and Nc are output information. (End of procedure W)

  FIG. 16 is an example of an explanatory diagram of the neighboring pixel search in the matching process. FIG. 16A shows the black pixel address A of the registered information of the registered fingerprint. When the registered fingerprint changed image black pixel address A and the test fingerprint changed image black pixel address A are completely the same, and when the test fingerprint changed image address A is not a black pixel, for example, the test fingerprint changed image address B is a black pixel. If there is, it indicates that the registered fingerprint changed image black pixel address A and the test fingerprint changed image address A are in close proximity. FIG. 16B shows the correlation between the perfect match and the neighborhood match. The black pixel of the test fingerprint changed image once referred to as the perfect match or the neighborhood match is the black pixel address of another registered fingerprint changed image. Indicates that the duplication check should be performed so as not to be treated as an exact match or a near match.

That is, the registered fingerprint changed image black pixel a is completely matched with the test fingerprint changed image black pixel d, the registered fingerprint changed image black pixel b is close match with the test fingerprint changed image black pixel e, and the registered fingerprint changed image black pixel The pixel c does not match the test fingerprint changed image, and it is determined that the already matched black pixel d or black pixel e is not coincidentally matched at the time of the perfect match check and the neighborhood match check. Is shown.
Note that the search range for determining the neighborhood match includes, for example, four neighborhoods and eight neighborhoods, and can be set in consideration of the state of image distortion at the time of input.

  Remark W (1): The value of Tci, (i = 1, 2,..., K) in step w3b is 0 ≦ Tci ≦ 1, and is determined as follows, for example. Assuming that the total number of black pixels of the registered fingerprint changed image is Nc, Nci / Nc, (i = 1, 2,..., K) represents the progress of processing, and the possible range is 0 ≦ Nci / Nc ≦ 1. It is. As Nci increases and approaches Nc, Nmi / Nci approaches Nm / Nc, which is the coincidence rate with {S, H, V} being examined at this time. Therefore, Tci increases as Nci increases. Can be set.

  As Tci increases, the range of abandonment in the middle increases, and the effect of reducing the amount of processing increases. On the other hand, erroneous recognition is more likely to occur, so an appropriate value needs to be set. The maximum value of the calculation is determined by the set value of k. If the abandonment in the middle of the collation is performed once, it is not necessary to calculate whether or not subsequent abandonment is possible for the value of {S, H, V} at that time. The specific numerical value of Tci needs to be determined depending on the characteristics of the target image. Also, the conditional expression that determines the range of abandonment in the middle of the collation is not limited to the example shown in the procedure W as long as it determines the degree of matching halfway.

Remark W (2): The expression of step W2 has the following meaning. In step W2, for all of the addresses (XR, YR), the new address after the parallel movement for making the approximate center point (XRC, YRC) of the registered fingerprint coincide with the approximate center point (XTC, YTC) of the test fingerprint is:
XR # = XR- (XRC-XTC)
YR # = YR-YRC-YTC)
And (XR #, YR #) is the new address. Next, rotation of the coordinate axis at an angle S about (XTC, YTC) is performed.

This means
XR ＠ = (XR # -XTC) · cos (S) + (YR # −YTC) · sin (S) + XTC = (XR−XRC) · cos (S) + (YR−YRC) · sin (S) + XTC
YR ＠ = − (XR # −XTC) · sin (S) + (YR # −YTC) · cos (S) + YTC = − (XR−XRC) · sin (S) + (YR−YRC) · cos (S ) + YTC
Can be obtained by Here, the values of the trigonometric functions sin (•) and COS (•) may be stored in advance in the range of the variation of the angle S.

(14) Flow of Registration Process and Matching Process FIG. 17 is an example of a schematic flowchart of the fingerprint registration process and the matching process. The registration processing is processing for registering fingerprint registration information in the memory 6 of the image processing apparatus 1. The matching process is a process of determining the coincidence between the test fingerprint and the registered fingerprint. The outline of the flow from the input of the fingerprint to the registration or collation is shown in the following procedure Z.

(Procedure Z)
Steps ZA1 to ZA5 are processes common to the registration process and the collation process.
Step ZA1:
The fingerprint is input from the image input device 2 to the image memory 4.
Step ZA2:
The fingerprint density image in the image 10 of the image memory 4 is smoothed.

Step ZA3:
The image 10 is subjected to binarization and background separation by a procedure B.
Here, after the binarization, smoothing and binarization may be performed on the binary image in order to reduce noise on the binary image.
Step ZA4: Find an approximate center point of the fingerprint image in the image 10. (End of step ZA1 to step ZA4)

Subsequent processing is divided into registration processing and collation processing.
Steps ZR <b> 1 and ZR <b> 2 correspond to the registration process, and register the registration information of the registered fingerprint in the memory 6.
Step ZR1:
A thinning process is performed on the registered fingerprint binary image (first image) in the image 10 in the fingerprint area to obtain a registered fingerprint changed image (first changed image).
Step ZR2:
Processing (procedure R) of the registered information of the registered fingerprint is performed. (End of step ZR1 and end of step ZR2)

Steps ZC1 and ZC2 are the case of the collation processing, and collate the registered fingerprint with the test fingerprint.
Step ZC1:
The test fingerprint binary image (second image) in the image 10 is subjected to black pixel ratio stabilization or narrowing processing to obtain a test fingerprint change image (second change image). If the black pixel ratio stabilization is performed in step ZA3, it is not necessary to perform the same repeatedly.
Step ZC2:
The matching between the registered fingerprint and the test fingerprint is determined by the matching process (procedure C, procedure W). (End of step ZC1 to step ZC2)

  In the image registration process, a plurality of images are sequentially input and held, and the images are collated, and an image having the highest matching rate can be set as a registered image. FIG. 18 is an example of a schematic flowchart of a registration process performed by selecting the best registered fingerprint from a plurality of input images. An example of the procedure is shown in procedure RR.

(Procedure RR)
Step RR1:
The fingerprint image is input to the image memory, and the input fingerprint image is stored in a file.
This operation is repeated a specified number of times.
(Here, a case where the specified number is three is described. The file names of the fingerprint images are, for example, z1, z2, and z3.)

Step RR2:
The input fingerprints are collated in the following combinations, and when the fingerprints match, the respective template coincidence rates T3 are stored in the memory.
If any one of them does not match, the registration process is terminated and the process is re-executed from step RR1.
The registered fingerprint z1 and the test fingerprint z2 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q12.

The registered fingerprint z1 and the test fingerprint z3 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q13.
The registered fingerprint z2 and the test fingerprint z1 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q21.
The registered fingerprint z2 and the test fingerprint z3 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q23.

The registered fingerprint z3 and the test fingerprint z1 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q31.
The registered fingerprint z3 and the test fingerprint z2 are collated, and when the final judgment is a match, the template matching rate T2 is obtained and set as Q32.
The number of final judgments must be equal to or greater than a specified condition (for example, when performing the matching process six times as described above, a specified condition such as four or more can be provided).

Next, an average matching rate for each input image is determined. For example, if everything matches,
As for z1, the average coincidence rate Ql is set to (Q12 + Q13) / 2.
For z2, the average coincidence rate Q2 = (Q21 + Q23) / 2.
For z3, the average coincidence rate Q3 = (Q31 + Q32) / 2.

Step RR3:
The maximum Qi among the average coincidence rates Qi = {Q1, Q2, Q3} is obtained, and registered with the registration information of the fingerprint zi corresponding to the i. (End of procedure RR)

  Here, the number of input images when registering one fingerprint can be set to one or more arbitrary values. Increasing the number of input images increases the possibility that a high-quality image can be selected as a registered fingerprint image, but on the other hand, the amount of registration processing increases, so the number of input images is determined in consideration of these factors. As for the selection of the combination in the collation between the input images, a part of the combinations may be selected instead of all the combinations as described above. The means for selecting the fingerprint image to be registered is not limited to the above-mentioned average coincidence rate, and another equation may be used.

(15) One-to-N Verification The verification in the above-described embodiment is to determine whether or not the registration information of one test fingerprint and one registered fingerprint match. In order to determine whether one test fingerprint matches at least one of the arbitrary number of registration information, it is necessary to determine whether one test fingerprint matches one registration information. Obviously, it is necessary to sequentially perform the processing on the information until the information matches. However, this method has a drawback that as the number of registration information increases, the amount of matching processing increases.
When it is sufficient to find one piece of registration information that matches one test fingerprint from an arbitrary number of pieces of registration information, the following procedure NA can be executed. FIG. 19 is an example of an explanatory diagram and a schematic flowchart relating to one-to-N matching. FIG. 19A shows the relationship between the test fingerprint and the registration information. A schematic flowchart is shown in FIG.

[Procedure NA]
The {S, H, V} positioning section is divided into divided ranges at the first, second,... Here, the original {S, H, V} range is the union of all the divided ranges. The division range may be set independently of the partial positioning section, or a partial positioning section may be used. As a simple example of the division range, the first time, a range that often coincides (for example, a partial alignment section of {S = 0, H = Hmin to Hmax, V = Vmin to Vmax}) is selected, The second time is to set a division range other than the first time.

(Procedure NA)
Step NA1:
Enter the test fingerprint.
Step NA2:
The division ranges are sequentially selected.
In the first matching with respect to the divided range, the matching between the test fingerprint and the registration information (i = 1, 2,..., N) is performed for the first {S, H, V} range. Repeat until the registration information can be found. If no matching registration information can be found from all target registration information in the first matching, the second matching is performed.

In the second matching with respect to the divided range, the matching between the test fingerprint and the registration information (i = 1, 2,..., N) is performed for the second {S, H, V} range. Repeat until the registration information can be found. If no matching registration information can be found from all target registration information in the second matching, the third matching is performed.
In the same manner, the matching process is performed until a matching registered fingerprint image is found. If a matching registered fingerprint image is found, the process ends. If a matching registered fingerprint image cannot be found, the process is performed up to the last divided range, and the processing ends. (End of procedure NA)
By this means, the registration information matching the test fingerprint can reduce the amount of collation processing until the matching registration information can be found.

(16) Substitution of Unmatched Part Ratio The following means can be used as a substitute for the determination of the mismatched part ratio described in the collation processing.
When comparing a binary image for registration with a binary image for inspection, registration information of the binary image for registration and registration information of an image obtained by reversing black and white of the binary image for registration (reversed registered image) Are registered separately, the matching between the binary image for inspection and the registered information is checked, and the matching between the inverted image of the binary image for inspection and the registered information of the inverted registered image is checked. When the final judgments of the two comparisons are both coincident, the binary image for registration and the binary image for inspection are judged to be identical, and the check based on the degree of inconsistency (for example, the inconsistency ratio) is omitted. I do. However, when this means is selected, the calculation of the unmatched portion ratio can be omitted, but there are disadvantages that the registration information increases and the collation time increases. The procedure of the processing related to registration is shown in procedure RVR. The procedure of the process related to the collation is shown in Procedure RVM.

(Procedure RVR)
Step RVR1:
The registration process of the registered fingerprint is performed, and the registration information is registered.
Step RVR2:
When the registered fingerprint is binarized, the white pixels and the black pixels are inverted, registration processing is performed on the binary image (referred to as an inverted binary registered image), and the inverted registration information of the registered fingerprint is registered. (End of procedure RVR)

(Procedure RVM)
Step RVM1:
The verification fingerprint is compared with the registered information. Here, the check of the unmatched portion rate is omitted.
Step RVM2:
When binarizing the test fingerprint, the white pixels and the black pixels are inverted, and the binary image (referred to as an inverted binary test image) is collated with the inverted registration information of the registered fingerprint. Here, the check of the unmatched portion rate is omitted. (End of procedure RVM)

  FIG. 20 is an example of a schematic flowchart of a registration process and a matching process when an inverted image is also used. FIG. 20A is a schematic flowchart of a process related to registration, and FIG. 4 shows a schematic flowchart of a process.

(17) Expansion or Modification The present invention is not limited to the above-described embodiment, and can be applied to, for example, the following expansion or modification. Image input method, smoothing processing, binarization processing, background separation processing, correction processing, processing to find approximate center points, narrowing processing, and formulas for calculating the matching rate and non-matching part rate in the matching processing However, the scope of the present invention is not limited to the present embodiment, and modifications, expansions, or partial omissions using other methods (for example, known methods) are possible. The setting method of the X coordinate and the Y coordinate is arbitrary. When the rotational deviation is so small as to be negligible, the alignment may be determined based on the positional deviation of only possible parallel movements, and may be determined based on the coincidence rate when the coincidence rate is the highest.

In step W2 of procedure W, the expressions for determining XR ＠ and YR ＠ can be used as long as they can perform rotation and translation, and are not limited to the present embodiment. For example,
XR ＠ = XR · cos (S) + YR · sin (S)
YR ＠ = − XR · sin (S) + YR · cos (S)
Can also be used. The use of coordinate transformation or geometric transformation is free. By adding, as registration information, values obtained by rotating or translating the sub-template, the processing amount for collation can be reduced (in this case, the memory amount increases).

Although the case where the image is a fingerprint has been described in the present embodiment, the present invention can be applied when the image can be regarded as being composed of lines. The sub-template and the non-sub-template can be freely divided, and there are extensions such as not dividing the two or providing more partitions.
In addition, the means described in the claims can be applied to an apparatus that selects one or more individually and performs arbitrary image processing.

It is a figure showing an example of 1 composition of a fingerprint identification system concerning one embodiment of the present invention. FIG. 3 is a diagram illustrating an example of use of an image memory and a memory. It is a figure showing an example of a pixel set. FIG. 9 is a diagram illustrating an example of division of a sub-template and a non-sub-template for a fingerprint area. It is a figure which shows the outline of the flow of the procedure IN which confirms input and validity of an image continuously, and fixes an image. FIG. 4 is a diagram illustrating an example of conversion of a processing range of an image. FIG. 9 is a diagram illustrating an example of division of an image into partial regions, and illustrating conversion of a luminance average value of the partial regions. It is a figure which shows an example of the outline | summary flow of the binarization process in the partial area | region where the brightness | luminance changes slightly. FIG. 11 is a diagram illustrating an example of an outline of a process for stabilizing a ratio of black pixels for binarization in a binary image divided into partial regions. FIG. 4 is a diagram illustrating an example of a fingerprint effective partial area table. FIG. 3 is a diagram illustrating an example of a hole (a set of white pixels) and a non-hole. FIG. 4 is a diagram illustrating an example of conversion of a binary image from a memory format to a compression format. It is a figure showing an example of the outline flow of the collation processing based on procedure C. FIG. 9 is a diagram illustrating an example of checking a mismatched part in a matching process. It is a figure which shows an example of the outline | summary flow of the image matching check assistance procedure (procedure W) in a collation process. FIG. 9 is a diagram illustrating an example of a neighboring pixel search in a matching process. FIG. 4 is a diagram illustrating an example of a schematic flow of a fingerprint registration process and a collation process. FIG. 9 is a diagram illustrating an example of a schematic flow of a registration process performed by selecting the best registered fingerprint from a plurality of input images. FIG. 2 is an explanatory diagram relating to one-to-N matching and a diagram illustrating an example of a schematic flow. FIG. 9 is a diagram illustrating an example of a schematic flow of a registration process and a collation process when a black-and-white inverted image is also used.

Explanation of reference numerals

DESCRIPTION OF SYMBOLS 1 ... Image processing apparatus, 2 ... Image input apparatus, 3 ... A / D converter, 4 ... Image memory, 5 ... Central processing unit (CPU), 6 ... Memory, 7 ... Imaging apparatus, 10, 11 ... Image, 12 ... Program and data, 13 ... Fingerprint registration information.

Claims (11)

For registered images and inspection images that are binary images,
Means for registering at least one address of a hole having a size within a specified range during execution of the registration processing means for the registered image;
During execution of the registered image and inspection image comparison processing means, for a specified range from the position of the registered hole address, when a hole exists in the inspection image, the hole is determined to match, and for all registered holes, Image processing, characterized by comprising means for examining and determining that the number of matched holes is equal to or greater than a specified ratio with respect to the total number of registered holes, a condition necessary for determining that the registered image and the inspection image match. apparatus. For registered images and inspection images that are binary images,
Means for registering at least one address of a non-hole, which is a portion where a hole of a specified range does not exist, during execution of the registration processing means;
During the execution of the registered image and inspection image comparison processing means, for the specified range from the registered non-hole address, when there is no non-hole in the inspection image, the non-hole is determined to match, and all registered non-holes are matched. Means that the registered image and the inspection image are required to determine that the registered non-hole is equal to or more than a specified ratio with respect to the total number of registered non-holes. Image processing device. A black pixel and a white pixel are components of a pixel of an image, and each pixel address is one of a pixel type of a black pixel and a white pixel as a target pixel type for a binary image represented by an X coordinate and a Y coordinate. ,
The position is arbitrary, a pixel set having a prescribed number of pixels and a shape, a type of a partial pixel set composed of a part of the pixel set is determined, and the pixel set is composed of a representative pixel and a non-representative pixel,
Regarding the binary image as a set of pixel sets, check whether the target pixel type exists in each pixel set for the X coordinate of the representative pixel of each pixel set on each Y coordinate, and determine whether the target pixel type exists. Only when the X pixel of the representative pixel, the division bit of whether the representative pixel is a black pixel or a white pixel, the division bit of a black pixel and a white pixel in each non-representative pixel, and the type data of a partial pixel set, An image processing apparatus comprising means for storing a binary image. For two images of a binary image in which a black pixel and a white pixel are constituent elements of the image pixels, a matching processing unit for checking the coincidence after aligning the two images,
By dividing a moving section in at least one direction into a plurality of partial sections in a moving range in the positioning, a partial positioning section is configured,
If the maximum value of the degree of matching in the first to m-th (m is a positive constant) partial alignment sections is out of the specified condition, the matching process is not performed in the subsequent partial alignment sections and the matching is performed. An image processing apparatus comprising: means for determining a mismatch. For two binary images having a black pixel and a white pixel as components of the image pixels, the matching processing means for checking the consistency after the alignment of the two images,
In a moving range in the positioning, a section of at least one direction of movement is divided into a plurality of partial sections to form a partial positioning section,
Means for sequentially performing collation processing for each of the divided partial registration sections, and when the final determination can be determined to be the first match, collation is not performed in the remaining partial registration sections, and the two images are determined to be coincident. An image processing apparatus comprising: For two binary images having a black pixel and a white pixel as components of the image pixels, the matching processing means for checking the consistency after the alignment of the two images,
In a moving range in the positioning, a section of at least one direction of movement is divided into a plurality of partial sections to form a partial positioning section,
When the set of target pixels of one image is successively separated by collation for one moving position in each partial registration section and sequentially checked for coincidence with the target pixel of the other image, each separated position is checked. Each time the check of the target pixel set is completed, the degree of matching in the middle is determined by the matching degree check condition for each of the separated target pixel sets. If there is no match, the collation at the moving position is mismatch. An image processing apparatus comprising: means for judging the movement position and abandoning the collation for the remaining portion of the movement position. For registered images and inspection images that are binary images,
Means for creating registration information from an image obtained by performing a narrowing process on a registration image;
Means for setting the ratio of the number of black pixels of the inspection image to the total number of pixels within black pixel ratio prescribed conditions,
Means for determining whether the degree of coincidence between the registration information and the inspection image is within the degree-of-coincidence prescribed condition,
Means for determining whether the degree of mismatch between the registration information and the inspection image is within the mismatch degree prescribed condition,
As the determination of the matching in the matching, it is determined that the degree of the match is within the condition for defining the degree of match and the degree of the mismatch is within the condition for defining the degree of mismatch.
An image processing apparatus comprising: means for setting conditions necessary to pass a match judgment between a registered image and an inspection image. For the inspection image and the registered image that are binary images, during the execution of the matching processing unit that compares the black pixels of the inspection image and the black pixels of the registered image,
When an arbitrary address A of the registered image is a black pixel and the address A of the inspection image is a white pixel, an address in a specified range near the address A of the inspection image is checked. The black pixel at address A is determined to be approximately the same as the black pixel in the inspection image, and in the subsequent comparison of the inspection image with the registered image, the black pixel determined to be approximately the same as the inspection image is registered. An image processing apparatus, comprising: a unit that excludes a black pixel other than an address A of an image from overlapping and judging a coincidence. A means for inputting and holding a plurality of images,
A means for comparing any two images,
Of the plurality of input images, two are sequentially selected in a prescribed combination, one of which is registered as a registered image, and the other is checked as an inspection image, and the degree of coincidence is stored. An image processing apparatus comprising: means for selecting a registered image when a determination can be made as a registered image from a plurality of input images.   When performing a matching process to find one registered image that matches one inspection image from an arbitrary number N (N ≧ 2) of registered images, a registration section between the registered image and the inspection image Indicates that the original registration section is separated so as to be a union of all division ranges, and that one inspection image is compared with N registered images for each division range of registration. An image processing apparatus comprising: means for sequentially performing one registered image that matches an image until the registered image can be found. For two binary images,
During execution of the registration processing means for the registered image,
Registration information of the registered image, and means for registering the registration information of the inverted registered image which is an image obtained by inverting the brightness of the registered image,
During the execution of the matching processing means,
Compare the inspection image with the registration information from the registered image,
The reversal test image, which is an image obtained by reversing the luminance of the test image, is compared with the registration information based on the reversal registration image,
If these two verification results are both passed,
An image processing apparatus, comprising: means for setting a condition necessary for determining that a registered image and an inspection image are the same.

Images