KR20060027613A - Direction filtering method in algorithm for extracting fingerprint minutiae - Google Patents

Abstract

The present invention is a direction filtering method for optimizing the filter used in the ridge interval scale region by reconstructing the image size of the input fingerprint image in the direction filtering using one size filter in the ridge interval scale region of the fingerprint image It is about. In the direction filtering method according to the present invention, a ridge interval of a fingerprint image, which can be generally released from a human, is classified by scale, set to a predetermined number of scale regions, and a scale region definition step of allocating a filter of one size for each scale region ( 100) and dividing the fingerprint image obtained from the person who wants the fingerprint recognition into blocks of a predetermined size and calculating an average value of the ridge direction and the ridge spacing in each block (200), and the average value and scale region of the ridge interval. Comparing the sizes of the filters allocated in the defining step to select a filter having a minimum difference (300), normalizing the fingerprint image to optimally match the size of the selected filter (400), and In step 500, direction filtering is performed on the image.

Fingerprint recognition, feature data, direction filtering

Description

Direction filtering method in algorithm for extracting fingerprint minutiae

1 is a flow chart of a conventional fingerprint feature data extraction method.

2 is a result of directional filtering fingerprints having narrow ridge spacing.

3 is a schematic diagram of a direction filtering method according to the present invention;

4 is a schematic diagram showing a method for calculating the ridge spacing.

The present invention is an improvement of the direction filtering step among fingerprint feature data extraction algorithms. In particular, the present invention provides an image of an input fingerprint image in the direction filtering using a filter of one size in the ridge interval scale region of the fingerprint image of the fingerprint image. The present invention relates to a direction filtering method of reconfiguring the size and optimizing the filter used in the ridge spacing scale region.

As one of security authentication methods using biometric information, a fingerprint recognition technique may be mentioned. Fingerprint recognition technology is a technology that determines whether or not a registered person by recognizing a fingerprint showing a unique characteristic difference for each person by an image processing method. In fingerprint recognition technology, the process of extracting fingerprint feature points and creating fingerprint feature data is the most important. An algorithm for generating fingerprint feature data is briefly described as follows.

First, a fingerprint image is obtained by reading a fingerprint by a fingerprint input device. The fingerprint image acquired by the fingerprint input device is divided into a plurality of areas. Next, the direction values of the fingerprint ridges are extracted for each region, and then the direction values of the ridges are binarized using a mask for each direction. After narrowing the width of the ridge to a single line, the feature points are found among the thinned ridges. After erroneously finding a feature point (pseudo feature point) among the found feature points, the feature data is calculated by making the position and the direction of the feature point data.

A conventional fingerprint feature extraction method will be described in more detail with reference to FIG. 1. After receiving the fingerprint image from the fingerprint input device (11), in the ridge direction extraction and correction step 12 (preprocessing step), the direction in which the brightness change in the entire fingerprint image area is the smallest is determined as the ridge direction of the area. For the correction of the ridge direction, the direction value of each region is determined in the direction averaged with the peripheral directions (also called direction correction, direction filtering, or smoothing). Next, binarization 13 is performed using this direction-specific information, followed by thinning 14 that converts the outline of the ridge into a valley until only the skeleton of the ridge exists. Finally, a known characteristic algorithm is used to extract disadvantages, branch points, intersection points, and the like which are inherent feature points of the fingerprint as a unique feature point for distinguishing the fingerprint (feature data extraction algorithm).

The preprocessing process is to extract better ridges and bones from the input fingerprint image, and is performed before the subsequent binarization. Since fingerprints are directional, they perform specific directional filtering or directional smoothing to obtain an improved image while preserving ridges.

The direction filtering step 12 is to increase the contrast of the ridges and valleys of the fingerprint image and reduce the noise before the binarization 13 and the thinning 14. However, if the ridge of the fingerprint is thinner or thicker than the standard, the direction filtering results in a distorted result from the original image. For example, when the ridge spacing of the fingerprint is narrow, adjacent ridges come together as shown in FIG. 2.

In fact, when applying the fingerprint recognition system, it is necessary to apply an appropriate filter size according to the ridge interval of the fingerprint to obtain an image suitable for the desired purpose. In other words, filtering should be performed by selecting an optimal filter according to the ridge spacing. This is because the ridge spacing has a certain pattern for the same person but different people may have different values in a certain area.

However, if the filter of the right size is selectively applied to each ridge interval during the direction filtering, the time cost for the calculation process needed to find and apply the appropriate filter, and the memory cost due to the various filter tables for the various filter applications, etc. Will increase.

In order to solve this problem, the fingerprint ridge spacing scale, which can be generally derived from humans, is modeled according to the probability distribution, and the whole is classified into a ridge-width scale space (hereinafter referred to as a "scale region"). A conventional technique has been developed in which a separate filter table is set for each scale region and a filter table having the same size assigned thereto is used within a predetermined region range, thereby performing filtering using the same filter in each ridge interval region. That is, in this prior art, the ridge interval is divided into predetermined ranges, and the optimum one is selected and applied among the filters set in advance for each range.

However, even in this type of filtering method, a fingerprint having a narrow ridge spacing may result in inappropriate filtering results. This is because one filter is applied to each predetermined scale region, and various fingerprint images must be filtered by one filter.

The present invention was developed to solve the above problems, based on the ridge interval scale area setting method as described above, but adjusting the size of the input fingerprint image filter size of the filter assigned to the scale area It provides a method for adjusting the image to fit the. If the filtering is performed after adjusting the input image, even if the ridge spacing is narrow, correct directional filtering can be performed while preserving structural features between the feature points.

As described above, the present invention comprises the following steps included in the direction filtering process of the algorithm for extracting feature data from a fingerprint image of a person who wants fingerprint recognition. In other words,

A scale region definition step of classifying ridge intervals of fingerprint images, which can be generally released from humans, by scale, setting them to a predetermined number of scale regions, and assigning filters of different sizes to each scale region.

Dividing a fingerprint image obtained from a person who wants fingerprint recognition into blocks of a predetermined size and calculating average values of ridge directions and ridge intervals in each block;

Selecting a filter corresponding to a region where the difference is minimal by comparing the average value of the ridge spacing with the sizes of the filters allocated in the scale region defining step;

Normalizing the fingerprint image to optimally match the size of the selected filter,

Performing direction filtering on the normalized fingerprint image using a selected filter.

Hereinafter, the configuration of the direction filtering method for fingerprint feature data according to the present invention will be described in detail. Referring to FIG. 3, in the direction filtering method according to the present invention, a ridge interval of a fingerprint image, which can generally come from a person, is classified by a scale, set as a predetermined number of scale regions, and a filter of one size is allocated to each scale region. A scale region defining step 100, dividing a fingerprint image acquired from a person who wants fingerprint recognition into blocks having a predetermined size, and calculating an average value of ridge directions and ridge intervals in each block (200); Comparing the average value of the interval and the size of the filters allocated in the scale region definition step (300) to select a filter with a minimum difference, and to optimize the size of the filter selected after the filter size selection step (300) Normalizing the fingerprint image to match (400), and performing a direction filtering on the normalized fingerprint image (500). It is.

The average value calculation step 200, the filter size selection step 300, and the fingerprint image normalization step 400 are particularly a direction filtering process during the preprocessing step 12 of an algorithm for extracting feature data from a fingerprint image of a person who wants fingerprint recognition. Included in the. That is, the present invention is an improvement of the direction filtering step corresponding to the preprocessing step of the existing fingerprint data extraction algorithm, which is technically included in the direction filtering (direction correction) step performed before the binarization step 13 shown in FIG. Will be.

First, the defining step 100 of the scale region based on the ridge model will be described. The definition of the scale region corresponds to the size of the filter to be applied to the fingerprint recognition system for each scale region. For example, the scale region of the ridge of the fingerprint image is classified into five, and each filter has a filter size of 3, 5, 7, 9, and 11, respectively. In this example, the number of filter sizes to be applied to the system is set to five, but of course the idea of the present invention is not limited to these five. Although the effect may vary, it is obvious that the spirit of the present invention is applicable to a system using only one filter. An object of the present invention is to normalize a fingerprint image in order to apply an optimal filter size to an input fingerprint image, which corresponds to a problem of best matching scale selection.

In the state where the scale region is defined as described above, as in the general fingerprint data extraction algorithm, the obtained fingerprint image is divided into a plurality of blocks to determine the ridge direction and the ridge interval. The method of dividing the acquired fingerprint image into a plurality of blocks is the same as the conventional method.

Next, the ridge interval is detected from the acquired fingerprint image. When looking at the signal of the cross section perpendicular to the direction of the ridge in the detection target window 30 (see Fig. 4) in any one block (see “20” in Fig. 4), the smoothing is performed to reduce the influence of noise on the image. The magnitude of the ridge spacing (a in Fig. 4) is found by detecting the value between the peak values of adjacent ridges in the ridge direction. After calculating the spacing values of adjacent ridges, their average representative values are calculated in the block.

Next, an average value for the entire block is obtained. This may utilize Equation 1 below. Equation 1 is an equation for calculating an average of ridge spacing values for all blocks of the acquired fingerprint image.

T : 영상영역(foreground region)에 존재하는 블록의 개수

T : the number of blocks in the foreground region

Next, it is selected to which scale region the processed fingerprint image corresponds. This is accomplished by selecting a scale factor (k) which minimizes the difference between the filter size and the scale region by using Equation 2 below as a step for selecting an optimal filter allocated for each scale region.

,

,

: 융선간격 스케일영역에 할당된 최적의 필터 사이즈

: Optimal filter size allocated to ridge spacing scale region

If the average filter size and the minimum filter size are selected through Equation 2, the direction filtering is performed using this filter. At this time, the fingerprint image is normally normalized to the selected filter size for more reliable filtering. That is, in image processing such as binarization of fingerprint images, optimal image quality can be obtained when the optimal filter size and average value of ridge spacing are the same, but it is impossible to design optimal filters individually for all fingerprint images. In turn, the fingerprint image is optimized for the filter size. This technical idea starts from which filter is selected from among the preset filters, and the filter selection method selects a filter which is the minimum of the ridge interval and the filter size (the average value of the ridge interval is small). Reconstruct the fingerprint image to match the filter size. In normalizing a fingerprint image, it is preferable to implement the present invention in a direction of selecting an optimal filter size by enlarging the input image since the input data is reduced when the input image is reduced.

This process is applied to the whole area using the average value of the ridge spacing except for the background part of the fingerprint image. This is because it is possible to obtain valid feature point information for the original input by knowing the scale parameter and the feature point extraction point. Image normalization may be performed by Equation 3 below.

여기서, : 스케일 팩터, I: 원 이미지, I': 스케일 조정된 새로운 이미지

Where: scale factor, I: original image, I ': new scaled image

According to the present invention, it is possible to improve the fingerprint recognition rate by efficiently and optimally directional filtering by coping with any type of input fingerprint image in the direction filtering in the way of assigning the filter for each scale region.

Claims (4)

In the direction filtering of an algorithm for extracting feature data from a fingerprint image of a person who wants fingerprint recognition, Scale region definition step of classifying ridge intervals of fingerprint images that can be generally released from humans by scale, setting them to a predetermined number of scale regions, and assigning filters of different sizes to each scale region, Dividing a fingerprint image obtained from a person who wants fingerprint recognition into blocks of a predetermined size and calculating average values of ridge directions and ridge intervals in each block; Selecting a filter corresponding to a region where the difference is minimal by comparing the average value of the ridge spacing with the sizes of the filters allocated in the scale region defining step; Normalizing the fingerprint image to optimally match the size of the selected filter, And performing a direction filtering on the normalized fingerprint image using a selected filter. The fingerprint feature data extraction algorithm according to claim 1, wherein in the filter selection step, the filter is selected by obtaining a value κ (scale factor) at which the difference between the filter size and the scale region is minimized by the following equation. Directional filtering method in.

,

: Optimal filter size allocated to ridge spacing scale region

The method of claim 1, wherein the filter selection step And selecting a filter having a minimum difference among the calculated average values of the ridge intervals smaller than the filter size. The method of claim 1, wherein the fingerprint image normalization is performed by the following equation.

here,

: Scale factor, I: original image, I ': new scaled image

Images