JPS59780A - Pattern collating device - Google Patents

Abstract

PURPOSE:To realize the high speed, by splitting a coordinate area of a characteristic point position into plural blocks and limiting the objective of check only to a characteristic point of an opposite pattern existing in a block near the block belonging to its characteristic point as to one characteristic point. CONSTITUTION:A pattern characteristic coded for each characteristic of a retrieval pattern and a file pattern are stored in a characteristic point memory 3. An operating circuit 7 performs the operation required for splitting and referencing of blocks with the indication of a control circuit 6 and the operated block and the characteristic number and the like are stored in a buffer memory 8. The coordinate area of the retrieval pattern and file pattern is split into plural blocks. Further, a proximity block to be retrieved is designated among the plural blocks in response to the position of each characteristic point of other patterns of the retrieval or the file pattern for attaining comparison and collation.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern matching method, and more particularly to a pattern matching method that automatically matches the identity of a striped pattern represented by a fingerprint using coded pattern features.

In an automated matching method for a striped pattern represented by a fingerprint, the special points of emphasis of the ridge pattern constituting the striped pattern, that is, the points M where the ridges break (end points) and the branching or merging points in FIG. The point m (branch point) is taken as a feature point of the pattern, and quantities (pattern features) representing the characteristics of each feature point including its position X, Y, and direction are coded and stored, and the coded pattern features are is used to compare and match search patterns to be identified, such as latent fingerprints left at crime scenes, with a large number of file patterns that have been registered and stored in advance on magnetic disks, etc. be exposed.

A method using the position X, Y, and direction of the feature point as the pattern feature described above is described, for example, in Japanese Patent Application Laid-Open No. 53-122353. In addition, we also calculate the difference between the "concentration amount" that indicates the density of feature points existing in the vicinity of each feature point, and the nearest neighbor feature point in each fan-shaped area divided into multiple areas around each feature point. A method using "relation" indicating the number of ridges as a pattern feature is disclosed in Japanese Patent Laid-Open No. 56-24675.

In either case, in the process of pattern matching, one feature point of the search pattern is compared with the coded pattern feature person including all the features of the file pattern at positions X and Y, and each of the two patterns is compared. Verification work is performed to sequentially check whether the difference in features is within a predetermined threshold. In such conventional methods, calculation comparisons are repeated for all combinations of feature points of the search pattern and feature points of the file pattern, resulting in a large number of calculations, especially when performing highly accurate matching. The drawback is that it takes time.

The purpose of the present invention is to divide the coordinate area of the feature point position (x, y) into a plurality of flocks, and for one feature point, the characteristics of the opponent pattern that exist in the block near the flock to which the feature point belongs. It is an object of the present invention to provide a high-speed pattern matching method that eliminates the above-mentioned entry by limiting the test target to points only.

The pattern matching method of the present invention compares and matches a striped search pattern to be identified and at least one pre-registered file pattern to be matched using coded pattern features of feature points of the pattern. In the matching method, a coordinate area is divided into a plurality of blocks, and all feature points of either the search pattern or the file pattern are assigned to one of the plurality of blocks according to their existing positions, and this information is used. Register in a storage means, designate neighboring blocks to be searched from among the plurality of blocks according to the position of each feature point of the other pattern of the search pattern or file pattern, and specify the feature points of the other pattern. The method includes means for comparing the coded pattern feature with a feature point of the one pattern registered in the neighboring block, and is configured by comparing the search pattern and the file pattern.

Next, the pattern matching method of the present invention will be explained in detail using an embodiment and with reference to the drawings.

When performing automatic matching of linear patterns such as fingerprints, it is first necessary to encode the features of the pattern, which can be done using, for example, the pattern feature extraction device described in Japanese Patent Application Laid-Open No. 55-138174. carried out by At this time, the position and direction of each minutiae are determined, but these may not always be the same even if the fingerprints are from the same person, depending on the conditions under which the fingerprints are taken or pressed onto the paper. In other words, since the coordinate system of the search pattern and the coordinate system of the file pattern are originally determined separately,
Prior to comparison and verification, it is necessary to make the coordinate systems the same. This method has already been established,
For example, the method described in JP-A-53-12235 is used.

FIG. 2 is a block diagram showing a specific example of dividing the Hatern coordinate area according to the present invention, and is a block diagram showing a specific example of dividing the Hatern coordinate area according to the present invention.
X=0 of the coordinate system standardized by the method described in Publication No. 235
．． 'The area coded from (-200) to (IFF) in 1C notation with Y=O as the center is divided into 80 parts (the 16 sections in the center are
(It is counted repeatedly as one section). The alphanumeric characters in each block are the block numbers, and the numbers shown in each part divided by the dotted line of the blocks 4C and 2C are the at quadrant numbers (r) of each quadrant obtained by further dividing each block into four. X' and Y' written together with the X and Y coordinates are position coordinates that have undergone coordinate conversion and are used to calculate the block number of each feature point, which will be explained later.

The main part of the present invention is, for example, a first step of registering all feature points of a file pattern converted into the same coordinate system into one of the plurality of flocks shown in FIG. 2 according to their positions; Determine the corresponding block and quadrant from the positions of the feature points of the search pattern converted to the same coordinate system, specify the blocks to be searched around the block with reference to the nearby block list shown in Figure 3, This includes a second step of sequentially outputting the feature points of the registered file patterns.

The neighboring block list shown in Fig. 3 is the 8 block list shown in Fig. 2.
Each of the 0-divided blocks registers the block number that touches each quadrant. For example, block OO
Quadrant number r=The block number adjacent to Q is 401411
44, the block number adjacent to r=1 is 01141, r=
The block numbers adjacent to r=2 are 04, 44, and the block numbers adjacent to r=3 are 01.04°05. In this list, EOD indicates an end symbol, and for example, FF is registered. Block number 40~4F
Since there are many blocks in contact with , up to 8 blocks can be registered.

FIG. 4 is a block diagram of an embodiment of hardware for carrying out the main part of the present invention described above, in which the coded pattern features of each feature point of the search pattern and file pattern are connected to data buses 1 and 2. It is stored in the feature point memory 3 via the feature point memory 3. The amount of coordinate transformation for making the coordinate systems of the search pattern and file pattern the same as described above is stored in the memory in the coordinate transformation circuit 4 via the same data bus. The coordinates regarding the positions and directions of the feature points stored in the feature point memory 3 are read out through the coordinate conversion circuit 4 without being converted. The control memory 5 is a memory that stores firmware (microprogram) necessary for a series of processes, and the control circuit 6 decodes this microprogram and executes the process. The arithmetic circuit 7 performs the computation necessary for block division reference based on instructions from the control circuit 6, and the computed block and feature point numbers etc. are used to store all feature points of the 3° file pattern stored in the buffer memory 8. The above-mentioned first stage process of registering in each block is performed according to the procedure of the registration flowchart shown in FIG.
This is done by creating two registration lists for P. That is, the positions X and Y of the feature point with feature point number f are read from the feature point memory 3, and the corresponding block number is
The operation of calculating BLKH and BLKP in the memory 8, calculated by the calculation circuit 7 according to the procedure shown in the figure, is repeated for all feature points starting from the feature point number (=Q. BLKH is the block number Tp and Hp, each consisting of 1 byte, are recorded for each block, where Tp is the minutiae number added to the last minutiae number registered in the block, and Hp is the minutiae number added to the minutiae number registered first. Indicates the number with 1 added. BLKP is a record of 1 byte Np for the feature point number f of the file pattern, and the feature point number f
The feature point number registered in the same block after ~1 is Np
-1, and Np is 001-1: EOI, (
The end symbol) indicates that f-1 is the last feature point number registered in that block. From these two lists, the blocks to which all the feature points of the file pattern belong can be found, and all the percentage points of the file pattern are registered in the corresponding blocks. Figure 7 shows the position coordinate values X of the signed 10 pits read out from the feature point memory,
A calculation flowchart for calculating a corresponding block number from Y, and X.

After converting Y to the X'', y''Km standard shown in Figure 2, we can calculate the 8-bit value by testing the exclusive OR of the upper 2 bits of the binary number X, Xl, and y. A flock number is calculated. X2, Y; and X''s, Y- are the upper 3rd and 4th bits of X'' and Y' respectively.
It is a binary number.

Next, the second step of determining the corresponding block and quadrant from the positions of the feature points of the search pattern and sequentially outputting the feature points of the file pattern to be searched will be described. In the first stage, all the feature points of the file pattern were registered in a block at once, but in this stage, the number of feature points of the file pattern to be searched for one search pattern feature point is determined according to the flowchart shown in Figure 8. Follow the steps to list and output as a list of nearby feature points. This is a general pattern matching algorithm, which searches for nearby feature points of the file pattern based on the feature points of the search pattern.
This is because after examining the pattern features including the arrangement relationship, the process moves to the next feature point of the search pattern. First, the positions X and Y of the feature point S of the search pattern are read out, and the block number in which the feature point S exists is calculated according to the procedure shown in FIG. Next, calculate the quadrant number r by further dividing each block into four, and refer to the neighboring block list (BLKR) in FIG. 3, which has already been written to the buffer memory 8 via the data bus 1, for the block number and r. to read the neighboring block number. The quadrant number r is the coordinate transformed position X:Y' in FIG.
Of the 10 binary numbers, each 2-bit binary number follows the bit used to indicate the block number (X; , X ; for small blocks.

Y',, Y'5. For coarse blocks X2 + X'3
+Y2rY:1). next,
Two registration lists BLKH in FIG. 6 are created based on the read neighboring block numbers.

The feature point numbers registered in each neighboring block are sequentially read from BLKP and omitted as neighboring feature point numbers f:A;...fi in the list of neighboring feature points shown in FIG.

At this time, the neighboring block number read from BLKR includes the number of the block in which the feature point S exists, and the neighboring feature point list does not include file pattern feature points in all directions near the feature point S. The first and second stages, which are the main parts of the present invention, have been described in detail, and the list of nearby feature points obtained in this way is stored in the buffer memory 8, The coded positions, directions, and other pattern features of the file pattern feature points registered in this list are sequentially read out to a pair of detection units (not shown in the block diagram in Figure 4) and used to detect the search pattern. The feature point S is compared with each pattern feature, and only those whose difference is smaller than a predetermined threshold value are selected as feature points that are one above. When the test for one search pattern feature point S is completed,
The procedure from the second stage described above is repeated for the next search pattern feature point S+1, and the test is performed for all search pattern feature points! When this is completed, the "pair" detection with one file pattern is completed, and the procedure from the first stage is repeated for the next file pattern. The thus-selected "pair" of minutiae points are compared in detail with each other's pattern features in a matching determination section, and the degree of pattern matching is determined. The circuit described in Japanese Unexamined Patent Application Publication No. 56-24675 can be used for the one-pair detection section and the matching determination section.

In the conventional method described in JP-A-56-24675,
In the "pair" detection section, all file pattern feature points within the area shown in Fig. 2 are compared with one search pattern feature point, whereas in the method of the present invention, 3 to 7 Since it is only necessary to test the file pattern feature points in the neighboring blocks, it is obvious that the number of calculation processes is greatly reduced, and the above-mentioned 1. Even if the second stage processing is taken into account, the processing time can be greatly reduced.

In the above embodiment, the block division is set to 80, and the peripheral part is coarsely divided and the central part is finely divided in consideration of the actual distribution of minutiae points of a fingerprint. However, the number of divisions and the division method are not limited to this. , the position of the search pattern is determined in each quadrant that further divides the above block into four, and the neighboring blocks are determined, but it does not necessarily have to be divided into quadrants, and therefore the neighboring block list is not limited to the format shown in Figure 3. . Further, the configuration of the computer is not limited to that of the embodiment, and the method of registering in each block and the format of the registration list are not limited to those of the embodiment. Further, in the embodiment, the feature points of the file pattern are examined using the feature points of the search pattern as a reference, but the processing can be performed using the feature points of the file pattern as a reference. ) Note that it is clear that the same effect can be obtained even if a pattern feature encoding device, a "pair" detection section, a matching determination section, etc. other than those shown in the example are used.

As explained in detail above, according to the pattern matching method of the present invention, the coordinate area of a pattern is divided into a plurality of blocks, all feature points of one pattern to be matched are registered in each block, and the coordinate area of the other pattern is registered in each block. Comparing and testing only the feature points and the feature points registered in their neighboring blocks has the effect of significantly reducing the number of calculations and speeding up pattern matching.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram of feature points of a striped pattern, FIG. 2 is a block division diagram showing a specific example of pattern area division of the present invention, and FIG. 3 is a specific example of a neighboring block list of the present invention. 4 is a block diagram of an embodiment of the software that executes the main part of the present invention; FIG. 5 is a registration flowchart showing an example of the procedure for registering feature point blocks in the present invention; FIG. 6 is a list format diagram showing a specific example of the registration list, FIG. 7 is a calculation flowchart showing an example of the block number calculation calculation of the present invention, and FIG. 8 is an example of the procedure for creating a list of neighboring feature points of the present invention. The flowchart, FIG. 9, is a list format diagram showing a specific example of a nearby feature point list. 1.2... Data bus, 3... Feature point memory, 4... Coordinate conversion circuit, 5...
Control memory, 6...control circuit, 7...arithmetic circuit, 8...buffer memory. Fig. 1 Fig. 2 Fig. 3 R Fig. 4 Fig. 5 Fig. 6 F3LKHBLKP Fig. 7 Fig. 8 Fig. 9 ■Kokaku Seal] ■ Possible ■

Claims (1)

[Claims] In a pattern matching method in which a striped search pattern to be identified and at least one pre-registered file pattern to be matched are compared and matched using coded pattern features of pattern feature points, multiple coordinate areas are allocating all the feature points of either the search pattern or the file pattern to one of the plurality of blocks according to their existing positions, and registering this information in a storage means;
Designate neighboring blocks to be searched from among the plurality of blocks according to the position of each feature point of the other pattern of the search pattern or file pattern, and register the feature points of the other pattern and the neighboring blocks. The coded no() with the feature points of the one pattern
A pattern matching method, comprising means for comparing turn characteristics, and matching the search pattern with a front file pattern.