JPS62118488A - Feature extracting device - Google Patents

Abstract

PURPOSE:To effectively detect a feature and to improve a reliability of a system by providing a gain control means for controlling variable density threshold according to a black line length or a white line length on respective scanning lines in a picture. CONSTITUTION:A picture input part 21 supplies the variable density value of a picture element on the respective scanning lines of the stored various density picture to a binarization part 22, the variable density value of the picture element is compared with a prescribed variable density threshold to perform the binarization of the picture element and fed to an AGC part 24. In the AGC part 24, the black line length or the white line length on the binarization scanning line is obtained, and compared with a previously stored line length threshold. When the black line length or the white line length is larger than the line length threshold, the variable density threshold is subtracted by a previously stored fixed value in the binarization part 22 and the picture element producing the respective line length is binarized again. A coordinate means 23 coordinates two-dimensionally based on a position relation of the black line or the white line on the respective scanning lines obtained by the binarization and the black line or the white line on the adjacent scanning line.

Description

[Detailed Description of the Invention] 1. Field of Industrial Application] The present invention relates to a feature extraction device, and particularly to a feature extraction device that extracts features of striped patterns such as 'tlR' and fingerprints of an object to be inspected in pattern recognition. The present invention relates to a 1,'1 feature extracting device.

[Conventional technology]

Conventional feature extraction devices are used in systems that identify individuals using fingerprints15.
In order to achieve extremely high reliability due to the characteristic of being unchanging throughout life, pattern recognition technology is used to check the identity of collected fingerprints and pre-registered fingerprints. For example, the method and device for eliminating
> -78+ 1; - Proposed in the specification, Japanese Patent Application No. 59-778, etc.

In such a fingerprint verification device, a fingerprint image stamped with ink on a paper surface, or a finger placed on a transparent surface such as glass, is detected by using the optical boundary change of the glass from this placement section. The fingerprint image obtained by
In+Ius1. After performing photoelectric conversion of the fingerprint pattern using an imaging device such as Rial Television), this photoelectric conversion process uses a feature extraction device that extracts the fingerprint pattern 1.1. The feature extraction device uses the 21st fingerprint as a fingerprint pattern feature.
, , /l, the position (X, Y), direction D, and relation of the point where the ridges 11 that make up the small fingerprint pattern break, that is, the end point 12, and the point where the ridges 11 merge or diverge, that is, the branching point 13. Rj(i=]~4) etc. have been proposed. Here, relation II i is the nearest feature point 1[1j (j-1 to 4) in each quadrant of the local coordinate system (ξ, η) with this as the origin, and the origin M The relation of the feature point M is R, = O, R2, respectively in Fig. 2.
=1°R3=2, R,4""4. In addition,
For details on these, see Japanese Patent Application Laid-open No. 1973 =-], 051
It is described in Publication No. 5.

[Problems to be Solved by the Invention: In the conventional feature extraction device described in 1, the input grayscale fingerprint image is thicker than usual due to uneven stamping pressure on the paper surface, with multiple ridges touching. Because the ridges may appear as ridges or become unclear due to variations in shading values due to individual differences in perspiration amount, in the 1'ffi conversion process for feature extraction, binarization using a constant shading threshold requires scanning. Line-1
The black run length or white run length at - becomes unstable, which greatly affects feature extraction performed by secondary association of each black run or white run on adjacent scanning lines.

The purpose of the present invention is to achieve 1.! It is an object of the present invention to provide a feature extraction device that solves the problems of the feature extraction device and performs a binarization process that adjusts the density threshold depending on the black run length or white run length of scanning line -1-.

[T 1 stage 1 for solving the problem The feature extraction device of the present invention is a 15' feature extraction device that extracts the feature of a striped pattern that an object to be inspected has. an image input means for converting and storing; and a 20-1 converter for binarizing the image by comparing the gray value of the 'dO light image image element stored in +fif with a predetermined 2〃8 light threshold V[]; gain control means for adjusting the grayscale threshold according to the black run length or white run length in the image;
The 111 run of - has a correspondence numbering means for two-dimensionally associating the white run with the positional relationship between the black run and the white run on the adjacent scanning line.

Furthermore, the feature extraction device of the present invention is a feature extraction device for extracting features of a striped pattern of an object to be inspected, and includes an image input means for photoelectrically converting a grayscale image of the object to be inspected and storing the image; A binarization means that binarizes the image by comparing the gradation value of the pixel in the gradation image with a predetermined gradation threshold; Alternatively, a noise removing means detects a white run, inverts the black run to a white run, and performs noise removal by inverting the white run to a black run; gain control means for adjusting the grayscale threshold according to the length or white run length; and means for performing two-dimensional correspondence based on positional relationships.

1. Effect]=6-Generally, the ridge width or the valley line width is approximately constant at the locations where the ridges flow in 11 lines on the fingerprint of the same finger. So 2. ,,,, 1 / I painting 1 Black run of each running II line in resentment. J(is 5)
In other words, a run with a predetermined ridge width (if it is longer, it is determined that multiple ridges are in contact), and a grayscale threshold is set for the pixel that caused the black run on scanning line 1-. , the valley line between the contacting ridges is detected by subtracting a predetermined constant 61 (.contrary to this, the white run length (here, the sequence length of 0'' of the 2 direct signal), that is, the valley If the line width is longer than the run length, there is a possibility that the ridge is not detected.
), and the d3 light threshold 6rE is added by a constant value of 11f to the pixel where a white run occurs on the scanning line to detect an unclear ridge. Here, it is assumed that the smaller the shading value, the darker the culm, and the larger the shading value, the whiter the culm.

Furthermore, when the ridges run in an oblique direction or in an arc shape, the ridge width of the scanning line 1- that crosses them is expected to be larger than that of normal ridges; : Even if the tl value is lowered, a valley line will not occur, and false detection of a valley line will not occur due to re-binarization. Also about the valley line.

Similarly, even if the gradation value is set to - when the valley line width is larger than the normal width, erroneous ridge detection will not occur.

1 Example 71 Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows a first embodiment of the present invention.

In FIG. 1, the first embodiment has an image input unit 21 in which a grayscale image of the object to be inspected is scanned pixel by pixel and photoelectrically converted, and then the grayscale value of each pixel is stored. . The image input unit 21 converts the grayscale values of pixels on each scanning line of the stored grayscale image into the binarization unit 22 f! The binarization unit 22 compares the gray value of the pixel with a predetermined gray value, binarizes the pixel, and performs automatic gain control (AGC).
The information is sent to Department 24. In the AGC unit 23, the black run length or white run length of each binarized scanning line is determined.
It is compared with a previously stored run length value.

Here, if the black run length or the white run length is greater than the run length value, the binarization section 22 adjusts the density value by a pre-stored fixed value to increase or decrease the black run length or the white run length. Binarization is performed again on the pixels that caused the run. The above two processes are repeated for all scanning lines in the grayscale image.

Note that feature extraction by two-dimensional correspondence of white runs or black runs on each scanning line is described in Japanese Patent Application No. 54-396, for example.
4.8 and Japanese Patent Application No. 55-1-38174.

FIG. 3 shows the signal waveform in the binary 1hi processing of the first embodiment of the present invention. In FIG. The solid lines indicate the gradation values in the gradation image from A to S.

In FIGS. 3(1) and 3(c), the broken lines represent the results of binarization, and in FIG. 3(f), the dashed-dotted line represents the grayscale threshold for re-binarization. In FIG. 3, signal waveform 31
Although there are two ridge lines from the direct entry point to point B, the gradation value of the valley line is lower than the gradation interval V1, so the ridge line is not detected. In FIG. 3(1)>, the black run lengths are compared and the black run length 1-AB from the direct entry to the 11th point B is greater than the run length value T+-. (C> As shown, the shading value is a constant value T. As a result of decreasing the shading value and binarizing it again, two ridge trees were detected. The shading image stored in this way The same process is performed on all the scanning lines in the middle, and the black runs or white runs of all scanning lines -L are two-dimensionally correlated based on their positional relationships, and the ridge pattern and valley line pattern are Detection is performed.

Note that the two-dimensional correspondence between the black run and the white run can be achieved by using a method of associating the ones whose center positions are closest to each other.

Figure 4 shows the flowchart of the binarization process + -1 in the first embodiment.
- does not show. In FIG. 4, in the binarization process of this embodiment, a grayscale image is input and stored by photoelectric conversion in step S1. In step S2, the gradation value of each pixel on each scanning line 1- in the gradation image is compared with a predetermined gradation/darkness value to be binarized, and the binary signal value and its pixel position are stored. , Next, in step S3, a black run or a white run is detected for each scanning line, and its length is calculated. The determined run length 1-5 is determined by the predetermined run length TL at step 87'I.
If the white run is long, /l! In addition to the pre-stored 3-tone threshold value T (I), re-binarization is performed in step S5 and new ridge detection is performed. Then, ternarization is performed again and valley line detection is performed for the new l:.The above processing is performed for one scanning line in step SO, and the t'+
, repeat for all scanning lines in step S7.

Note that the number of re-binarization cycles 3j7L for each black run or each white run on each scanning line is not limited to one time, but is based on the adjustment numerator of the grayscale value. or a predetermined run length TL
This can be done multiple times by changing the .

FIG. 5 shows a block diagram illustrating a second embodiment of the invention. In FIG. 5, the second embodiment is obtained by adding a noise removal section 55 to the first embodiment shown in FIG. In the second embodiment, in the same way as in FIG. . At this time,
If the run length is smaller than a pre-stored noise run length LN, the black run inverts its binary signal value to the white run, and the white run performs the same inversion to the black run to remove the noise. It is something to do. Since the other components are the same as those in the first embodiment, detailed explanation thereof will be omitted.

Figure 0 shows the flowchart 1 of the binarization process of the second embodiment.
Not shown. In FIG. 6, the processing in steps 81 to S3 is the same as in FIG. 3, and when the obtained black run length or white run length is smaller than the noise run length LN, the binary signal value is inverted, Noise is removed in step S41. Steps 351 to S8] are the same as steps 84 to S7 in the second embodiment.

FIG. 7 does not show a schematic diagram of the binarization process of the second embodiment.

In FIG. 7, from point 13 in FIG. 7(a) to point C
Until now, a pulse-like noise was produced by sweat gland pores, etc.
In Figure 7 (1)), the point! ] to point C is compared with the run length 'rT run length' [8, and as a result, the noise is removed as shown in the seventh country (c) 1, the 71st M ( d) is the same process as in FIG. 3(c).

〔Effect of the invention〕

As explained above, the present invention provides an apparatus for extracting the features of a striped pattern using pattern recognition technology, by locally changing the density threshold during binarization. Features such as points can be effectively extracted, and the performance of subsequent verification processing can also be improved, which has the effect of increasing the reliability of the system.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a first embodiment of the present invention;
3 is a schematic diagram showing the operation of the first embodiment, FIG. 4 is a diagram illustrating the flowchart 1 of the first embodiment, and FIG. 5 is a diagram explaining the characteristics of a fingerprint pattern. FIG. 6 is a block diagram showing a second embodiment of the present invention, FIG. 6 is a flowchart of the second embodiment, and No. 71:A is a schematic diagram showing the operation of the second embodiment. 11... Ridge, 12... End point, 13... Branch point,
21゜51... Image input section, 22.52... Binarization section, 23゜53... A G C section, 24.54...
Correspondence is in section 55...noise removal section. Treetops 3 figures 4 figures almost 7 figures

Claims (2)

[Claims] (1) In a feature extraction device for extracting features of a striped pattern of an object to be inspected, an image input means for photoelectrically converting and storing a grayscale image of the object to be inspected; A binarization means that binarizes the image by comparing the gradation value of the pixel with a predetermined gradation threshold, and adjusts the gradation threshold according to the black run length or white run length on each scanning line in the image. a gain control means, and a correspondence means for performing two-dimensional correspondence based on the positional relationship between the black run or white run on each scanning line obtained in the binarization and the black run or white run on an adjacent scanning line. A feature extraction device is characterized in that it performs feature extraction on a striped pattern. (2) In a feature extraction device for extracting features of a striped pattern of an object to be inspected, an image input means for photoelectrically converting and storing a grayscale image of the object to be inspected; A binarization means that binarizes the image by comparing the gray value of the pixel with a predetermined gray threshold; a noise removing means for removing noise by inverting the black run to a white run and inverting the white run to a black run; and a grayscale value according to the black run length or white run length on each scanning line in the image. gain control means for adjusting or subtracting the value; and means for two-dimensionally associating a black run or white run on each scanning line obtained in the binarization with a black run or white run on an adjacent scanning line based on a positional relationship. A feature extraction device characterized by extracting features on a striped pattern.