KR20210082624A - Fingerprint Enhancement method - Google Patents

Abstract

An objective of the present invention is to improve the quality of a fingerprint input to an online fingerprint identification system (automatic fingerprint identification system). The present invention relates to a fingerprint enhancement method. More specifically, the fingerprint enhancement method according to one embodiment of the present invention includes a Gabor filter bank application step, a ridge extraction stage, a Coles level ridge map and unrecoverable area mask generation step, and a direction field estimation step.

Description

Fingerprint Enhancement method

The present invention improves the clarity of the ridge/valley structure in the recoverable area and makes it suitable for the minutiae extraction algorithm, in order to improve the quality of the fingerprint input to the online fingerprint identification system (fingerprint automatic identification system), and also the actual ridge It relates to a fingerprint enhancement method that identifies and masks all areas of /valley structure that are damaged beyond repair, labeling them as non-recoverable areas and masking them.

In general, each fingerprint image has a different quality. Integrating a fingerprint enhancement module into an Automated Fingerprint Identification System (AFIS) system is essential to ensure that the performance of the Automated Fingerprint Identification System (AFIS) is robust with respect to the quality of the input fingerprint image.

The Automated Fingerprint Identification System stores personal information and the original image or key characteristic information of an individual in a high-speed, large-capacity database, and compares the individual's on-site fingerprint with the stored fingerprint in a search terminal when necessary. check whether It is mainly used for forensic investigation and immigration control.

In general, Automatic Fingerprint Identification Systems (AFIS) compare the fine details of the ridge/valley structure of the fingerprint, a total of 18 types of regional ridge/valley descriptions identified. Among them, ridge endings and ridge branches, commonly referred to as 'minutiae' (minimum, feature), are the two most prominent structures used in automatic fingerprint recognition systems. Extracting reliable minutiae automatically from digital fingerprint images is a very difficult task. The performance of currently available minutiae extraction algorithms is highly dependent on the quality of the input digital fingerprint image. The minutiae extraction algorithm generally assumes that digital fingerprint images may not always have a well-defined ridge/valley structure due to the abnormal formation of various factors (fingerprints, postpartum marks, occupation marks, acquisition device problems, etc.). This, of course, leads to the failure of the minutiae extraction algorithm, which degrades the performance of the fingerprint matching module. 1 and 2 show a typical example of applying a minutiae extraction algorithm to a fingerprint image of good or poor quality obtained by a conventional inkless fingerprint scanner. 1 and 2, (a) shows the input image, (b) shows the extracted ridge map, (c) shows that the extracted minutiae are superimposed on the input fingerprint image.

In order to ensure that the performance of the automatic fingerprint recognition system is robust with respect to the quality of the input fingerprint image, an improved algorithm that improves the sharpness of the ridge/valley structure is needed. Ideally, the ridge/valley structure of the fingerprint image is well defined. Each ridge is divided into two parallel narrow ridges, and each valley is divided into two parallel narrow ridges, and the minutiae is defined as the ridge end and the ridge branch. However, in practice, well-defined ridge/valley structures are not always displayed in scanned fingerprint images. In general, fingerprint images are damaged by various kinds of noise such as wrinkles, stains, and holes. Despite the presence of these noises, trained fingerprint specialists are often able to correctly identify minutiae using various visual cues such as local ridge/valley orientation, ridge/valley continuity, etc. Therefore, an algorithm that can rely on these visual cues to improve the quality of the input fingerprint image is desired.

In general, in the case of a designated fingerprint image, for fingerprint image recovery, it can be roughly divided into three categories of areas of interest.

The first category is a well-defined region where ridges and valleys are clearly visible, where the minutiae extraction algorithm can work reliably.

The second category is recoverable damaged areas where ridges and valleys are damaged by small amounts of wrinkles, stains, etc., where enhancement algorithms can properly repair them.

A third category is irreparable damaged areas, where ridges and valleys are damaged by a significant amount of noise and distortion that cannot be repaired in the image.

The first two categories of the noisy fingerprint area, that is, the first category and the second category, are recoverable, and the last category, that is, the third category, is unrecoverable. Efforts to improve the quality of fingerprint images in these areas are futile because it is impossible to recover true ridge/valley structures in unrecoverable image areas. Therefore, the goal of a reasonable enhancement algorithm is to improve the sharpness of the ridge/valley structure of the fingerprint image in the recoverable area and to mask the non-recoverable area. Another very important aspect regarding the fingerprint enhancement algorithm is that no spurious ridge/valley structures should occur.

Conventionally, various techniques have been proposed to improve the fingerprint image, and these techniques can adaptively improve the quality of the input fingerprint image by utilizing information about the local ridge/valley structure. However, all of these techniques assume that the local ridge/valley direction can be reliably estimated from the input fingerprint image. In practice, this assumption does not apply to poor quality fingerprint images.

Figure 3 shows some examples of expected orientation fields of poor quality fingerprint images. In the fingerprint enhancement algorithm, reliable computation of the direction field is a key problem.

Therefore, it is necessary to improve the sharpness of the ridge/valley structure in the recoverable region and make it suitable for the minutiae extraction algorithm. In addition, it is necessary to identify all damaged areas for which the actual ridge/valley structure cannot be repaired and label them as non-recoverable areas.

The present invention improves the sharpness of the ridge/valley structure in the recoverable area and makes it suitable for the minutiae extraction algorithm, and also identifies all damaged areas where the actual ridge/valley structure cannot be repaired and labels it as the non-recoverable area. To designate, a fingerprint enhancement module and a fingerprint enhancement method are proposed.

As a prior art, Korean Patent Laid-Open Publication No. 10-2007-0023217 relates to a method and apparatus for estimating a direction. In the case of this invention, since a coarse level ridge map and an unrecoverable area mask are not generated, the accuracy may be somewhat reduced due to unrecoverable areas in the fingerprint image.

As another prior art, Chinese Laid-Open Patent Publication No. 107871312 relates to a Gabor filter-based fingerprint improvement system technology. Even in the case of this invention, since a coarse level ridge map and an unrecoverable area mask are not generated, the accuracy may be somewhat reduced due to unrecoverable areas in the fingerprint image.

4 is a flowchart schematically explaining the fingerprint recognition step in Korean Patent Laid-Open No. 10-2017-0136370, a data capture step (S110), a fingerprint detection and feature extraction step (S150), a fingerprint matching step ( 200) is included.

Here, the data capturing step (S110) is a step of storing the user's fingerprint information in a memory unit (not shown) at the initial stage of use, and the fingerprint detection and feature extraction step (S150) is a data capture step (S110) pre-stored. It is a step for extracting the fingerprint information (feature points) of the fingerprint for comparison with the user's fingerprint information, and the fingerprint matching step 200 uses a feature amount defined from the extracted feature point information to obtain two fingerprint images ( That is, the degree of similarity between the fingerprint image of the data capture step (S110) and the fingerprint image of the fingerprint detection and feature extraction step (S150) is determined.

In the data capture step (S110) and the fingerprint detection and feature extraction step (S150), the fingerprint sensor unit (fingerprint scanner) (not shown) receives the fingerprint image at each moment in multiple layers and matches the image into one image, A feature (minutia) extraction unit (not shown) extracts a feature (minutia) from the image-matched fingerprint image.

In the present invention, in the data capture step (S110) and the fingerprint detection and feature extraction step (S150), a fingerprint enhancement module (not shown) is provided between the fingerprint sensor unit (not shown) and the feature extraction unit (not shown) In addition, it is proposed to improve the quality of the image-matched (that is, scanned) fingerprint image in the fingerprint sensor unit and transmit it to the minutia extraction unit (not shown).

The technical problem to be solved by the present invention is, in order to improve the quality of a fingerprint input to an online fingerprint identification system (automatic fingerprint identification system), it is decomposed into a filtered image set before input to the online fingerprint identification system, It is to provide a fingerprint enhancement method by estimating a directional field from a fingerprint image and generating a quality mask that distinguishes between recoverable and non-recoverable damaged areas.

Another technical problem to be solved by the present invention is to improve the clarity of the ridge/valley structure in the recoverable area and make it suitable for the minutiae extraction algorithm, and also identify all damaged areas where the actual ridge/valley structure cannot be repaired, and The goal is to provide a fingerprint enhancement method that labels (i.e. masks) the non-recoverable area.

Another technical problem to be solved by the present invention is that a uniform symmetric Gabor filter bank is applied to an input fingerprint image to generate a filtered image set, and a ridge extraction algorithm is applied to each filtered image to generate a corresponding ridge map. Then, using a voting algorithm on the extracted ridge map of the filtered image, a coarse-level ridge map and an unrecoverable area mask are generated, and the direction is given to the generated coarse-level ridge map. An object of the present invention is to provide a fingerprint enhancement method that obtains the local direction of each pixel by applying an estimation algorithm.

In order to solve the above problems, the fingerprint enhancement method of the present invention, the Gabor filter bank unit, using a Gabor filter bank, filters the input fingerprint image, generating a filtered fingerprint image, Gabor filter bank application step ; A ridge extraction step of generating a ridge map by extracting a ridge line from each filtered fingerprint image output in the Gabor filter bank application step; The quality mask generator applies a voting algorithm to the ridge map generated in the ridge extraction step to generate a coarse-level ridge map and an unrecoverable area mask, a coarse-level ridge map and recoverable generating a mask of an absent region; The direction field estimation unit includes a direction field estimation step of obtaining a local direction of each pixel from the coarse level ridge map and the coarse level ridge map generated in the unrecoverable region mask generating step. .

The ridge extraction step includes: a local direction field estimation step in which the ridge line extraction unit obtains a local direction field from the fingerprint image filtered in the Gabor filter bank application step; _{The ridge extraction unit obtains the consistency level (C 0} (u,v)) of the direction field obtained in the local direction field estimation step in the local area of the pixel (u, v), but the consistency level is a preset consistency threshold calculating the consistency level of the direction field, obtaining a local direction field exceeding the value Tc; an adaptive filtering step of applying, by the ridge line extraction unit, two masks composed of an adaptive filter to the fingerprint image filtered in the Gabor filter bank application step after the coherence level calculation step of the direction field; When the gray level value of the pixels (u, v) of the image to which the two masks of the adaptive filtering step are applied is greater than the _{preset ridge threshold value (T ridge ),} Comparing the gray level value and the ridge line threshold value in which the area is calculated by the ridge line extraction unit; The ridge line extraction unit determines whether the area calculated in the step of comparing the gray level value and the ridge threshold value _{is smaller than the minimum threshold value (T min} ). If it is small, it is designated as a background label. Comparing the area of each component of the ridge map and the minimum threshold value by dividing the ? into a set of line segments; The ridge extractor determines whether each line segment divided in the area of each component of the ridge map and the minimum threshold value comparison step is between a pair of parallel ridges, and if each line segment is between a pair of parallel ridges, it is an actual ridge line. and determining whether the line segment is positioned between the pair of parallel ridges by designating a label, and designating a label as a background label if each line segment is not between the pair of parallel ridges.

(단, θ _i와 θ _j는 특징(minutiae) i와 j가 고정된 참조 축에 종속된 각도이며, B는 선의 평균 밝기이며, D는 특징(minutiae) i와 j사이의 거리임)을 이용하여 후처리부가 가짜 소수를 표시하는 후처리단계;를 더 포함한다.In addition, the fingerprint enhancement method is a specific objective function (E)

(where θ _i and θ _j are the angles at which the features i and j are dependent on the fixed reference axis, B is the average brightness of the line, and D is the distance between the features i and j) and a post-processing step in which the post-processing unit displays a fake prime number.

In the Gabor filter bank step, after the Gabor filter bank unit performs FFT on the input fingerprint image to obtain a frequency image, the corresponding Gabor filter with a predetermined radial and directional frequency is applied to the frequency image, and the Gabor filter is applied to the frequency image. , an inverse FFT is performed to obtain a filtered image.

In the Gabor filter, the center frequency is 60 cycles/width (height), the radial bandwidth is 2.5 octaves, and _{the eight values of the center direction θ 0} are 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees. , 135 degrees, 157.5 degrees, and the direction bandwidth is 35 degrees.

In the adaptive filtering step, the two masks (h _t (u,v;i,j) and h _b (u,v;i,j)) composed of the adaptive filter are

(however, θ (u, v) indicates the direction of the local ridge line of the pixel (u, v)).

In the Coles level ridge map and unrecoverable region mask generation step, the quality mask generation unit divides each ridge map of the filtered image output from the ridge extraction unit into blocks of size 8 x 8, and If the number of ridge pixels is greater than or equal to a predetermined threshold number of ridge pixels, each block is labeled as foreground (value 1), and if the number of ridge pixels is less than the threshold number of ridge pixels, each block is set as a background (background) It is designated as a label (value 0), and the coarse level ridge map is applied according to the preset coarse level ridge map calculation rule, and the coarse level ridge map and the unrecoverable area mask creates

The coarse level ridge map calculation rule is that if only one of 8 binary block maps of pixel (x, y) has value 1 and this pixel belongs to a connected component of size K, then K is the block threshold ( T _block ), the corresponding block pixel value in the coarse level ridge map is duplicated in the associated ridge map, and the pixel value of the corresponding recoverable area mask is set to value 0 to indicate that the block can be recovered. , Rule 1; If two or more binary block maps at a pixel (x, y) have vlaue 1 and the connected local ridge directions are not orthogonal to each other, the pixel value of the corresponding block in the coarse level ridge map is regarded as the average value of the associated ridge map values. a second rule, wherein the pixel value of the corresponding recoverable region mask is set to value 0 to indicate that the block can be recovered; If two or more binary block maps at a pixel (x, y) have a value of 1, the associated local ridge directions can be orthogonal to each other, provided that the connected component has only one pixel whose _{size is greater than the block threshold (T block ).} , the third rule, in which the pixel value of the block in the coarse level ridge map is duplicated in the ridge map associated with the largest connected component and the pixel value of the corresponding recoverable area mask is set to value 0 to indicate that it is a recoverable block. ; If the first to third rules are not met, a fourth rule is included in which a value of 1 is assigned as a label indicating that the block cannot be recovered.

In the direction field estimation step, the direction field estimator obtains the local direction of each pixel on the coarse level ridge map, and then the gray level value (g(x,y)) of the pixel (x, y) of the improved image of

(where θ(x, y) represents the local direction field value of the pixel (x, y)).

In order to improve the quality of a fingerprint input to an online fingerprint identification system (automatic fingerprint identification system), the fingerprint enhancement method of the present invention decomposes it into a filtered image set before input to the online fingerprint identification system, but the filtered fingerprint image A directional field is estimated from , and a quality mask is generated that distinguishes between recoverable and non-recoverable damaged areas. That is, improve the sharpness of the ridge/valley structure in the recoverable area and make it suitable for the minutiae extraction algorithm, and also identify all damaged areas where the actual ridge/valley structure cannot be repaired and label it as the non-recoverable area. (i.e., perform a mask). By doing so, the accuracy is increased during authentication, and in particular, false ridge and valley structures are not generated.

In addition, in the present invention, a uniform symmetric Gabor filter bank is applied to an input fingerprint image to generate a filtered image set, and a ridge extraction algorithm is applied to each filtered image to obtain a corresponding ridge map, and the filtered image Generate a coarse-level ridge map and an unrecoverable area mask using a voting algorithm from the extracted ridge map of , and apply a direction estimation algorithm to the generated coarse-level ridge map. We provide a fingerprint enhancement method to obtain the local orientation of each pixel.

1 shows an example of a result of applying a minimum extraction algorithm to a high-quality fingerprint image obtained with an inkless fingerprint scanner.
2 shows an example of a result of applying a minimum extraction algorithm to a low-quality fingerprint image obtained with an inkless fingerprint scanner.
3 is an example showing an expected direction field of poor quality fingerprint images superimposed on an input image.
4 is a flowchart schematically illustrating a fingerprint recognition step in Korean Patent Laid-Open No. 10-2017-0136370.
5 is a schematic flowchart of a method for improving fingerprint quality according to the present invention.
8 is a result of applying the improved algorithm of the present invention to a fingerprint image with poor quality.

Hereinafter, a fingerprint enhancement module and a fingerprint enhancement method according to the perspective of the present invention will be described in detail with reference to the accompanying drawings.

5 shows a schematic configuration of a fingerprint quality improvement module of the present invention, and the fingerprint quality improvement module 10 includes a Gabor filter bank unit 20, a ridge line extraction unit 50, and a quality mask generation unit 60. , a direction field estimation unit 70 , and a post-processing unit 80 . Here, the fingerprint quality improvement module 10 may be formed of an arithmetic processing unit, that is, a computer or a microprocessor.

Gabor filter bank unit 20, including a uniform symmetric Gabor (Gabor) filter bank, filters the input fingerprint image to generate a filtered image set.

The ridgeline extraction unit 50 obtains a ridgeline map by applying a ridgeline extraction algorithm to each filtered image output from the Gabor filter bank unit 20 .

The quality mask generating unit 60 is a means for generating a coarse level ridge map and an unrecoverable region mask, and applies a voting algorithm to the ridge map output from the ridge extraction unit 50. Creates a ridge map with coarse levels and a mask of unrecoverable areas.

The direction field estimator 70 obtains a local direction of each pixel by applying a direction estimation algorithm to the coarse-level ridge map generated by the quality mask generator 60 .

Here, the Gabor filter bank unit 20, the ridge line extraction unit 50, the quality mask generation unit 60, and the direction field estimation unit 70 are means for processing the algorithm for improving the fingerprint, and the post-processing unit ( 80) is a means of dealing with postpartum cracks that cannot be treated by the enhancement algorithm.

Cracks in the fingerprint image can be characterized as narrow bright areas running transversely in the direction of the generally dominant ridge. One of the inferences for detecting minutiae due to these cracks is based on the observation that the fractional terms are semi-aligned and the region between them is brighter than the average brightness of the foreground region.

The post-processing unit 80 displays a fake prime number using a specific target function. That is, from the input fingerprint image and the set of primes detected in the enhanced image, information necessary to identify these primes is obtained.

6 is a schematic flowchart of a method for improving fingerprint quality according to the present invention, and FIG. 7 is a flowchart schematically illustrating a ridge extraction process in the ridge extraction step S320 of FIG. 6 .

Gabor (Gabor) filter bank application step (S310), a uniform symmetric Gabor (Gabor) filter bank (ie, even symmetric Gabor filters) in the Gabor filter bank unit 20 is applied to the input fingerprint image and the filtered image set is created

A fingerprint is a fluid pattern composed of locally parallel ridges and valleys. A human fingerprint has a local frequency and a local direction. A band pass filter can efficiently remove unwanted noise and preserve the true ridge/valley structure. The Gabor filter has both frequency-selective and direction-selective characteristics, and has optimal joint resolution in both spatial and frequency domains. Therefore, it is recommended to use a Gabor filter as a band pass filter to remove noise and preserve the true ridge/valley structure.

Here, the Gabor filter is a type of linear filter, and its impulse response is a product of a harmonic function and a Gaussian function. Because of the convolution property, the Fourier transform of the impulse response of the Gabor filter is the convolution of the Fourier transform of the harmonic function and the Fourier transform of the Gaussian function. In general, a Gabor filter is a filter that detects an outline, and in simple terms, it can be referred to as a Gaussian filter modulated by a sine function. As the size or direction of the edge can be changed by adjusting the parameter, it is applied to the feature point extraction algorithm.

The general form of an even symmetric Gabor filter is:

where h(x,y) is the transfer function of the Gabor filter, u ₀ is the sinusoidal plane wave frequency along the x-axis, and δ _x and δ _y are the spatial constants of the Gaussian envelope along the x and y axes, respectively. . Gabor filters with arbitrary orientation can be obtained through rotation of the xy coordinate system. The modulation transfer function (MTF) of the Gabor filter is as follows.

where H(u, v) denotes the modulation transfer function of the Gabor filter, δ _u is 1/2πδ _x (i.e., δ _u = 1/2πδ _x ), δ _v is 1/2πδ _y (i.e. δ _v = 1/2πδ _y ).

An important issue when applying Gabor filters is the selection of filter parameters. In a fingerprint image of size 512 x 512, the ridge frequency is typically about 60 cycles per image width (height). Therefore, in the fingerprint enhancement algorithm, the center frequency is selected as 60 cycles/width (height). The radial bandwidth is chosen to be 2.5 octaves. Eight values of the central direction θ ₀ are used: 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees, 135 degrees, and 157.5 degrees. The direction bandwidth is chosen to be 35 degrees. For a given input fingerprint image, these 8 Gabor filters are applied to obtain 8 filtered images. To obtain the filtered image, FFT is first performed on the input fingerprint image, then the corresponding Gabor filters with adjusted radial and directional frequencies are applied to the frequency image and an inverse FFT is performed to obtain the filtered image.

In the ridge extraction step S320 , the ridge extraction algorithm is applied to each image filtered in the Gabor filter bank application step in the ridge extraction unit 50 , and the corresponding ridge map is extracted to the filtered image. This ridge map is used to generate a coarse-level ridge map and a coarse-level ridge map of the input fingerprint image in the step of generating an unrecoverable region mask. The most prominent property corresponding to a ridge in the filtered image is that the gray level value of the ridge achieves a local maximum along a direction orthogonal to the local ridge. Pixels can be reliably identified as ridged pixels based on this property.

A ridge line extraction method (ie, a ridge line extraction algorithm) will be described below with reference to FIG. 7 .

As the local direction field estimation step (S321), the ridge line extraction unit 50 is applied to each image filtered in the Gabor filter bank application step, or to an image whose resolution is lowered in the comparison step of the consistency level and the consistency threshold value. , apply the following formula to estimate the local direction field.

where W is the size of the local window centered on pixels (u, v), which can be done with W=15 in this algorithm. Gx and Gy are the gradient magnitudes in the x and y directions, respectively, and Δ _x (u, v) and Δ _y (u, v) are the values of the estimated vector field of the input fingerprint image at a given pixel (u, v). indicates This operation computes the dominant direction of the Fourier spectrum of the local window around pixel (u, v).

In the direction field consistency level calculation step (S322), the ridgeline extractor 50 calculates the consistency level (C ₀ (u, v)) of the direction field in the local area of the pixel (u, v) by the following equation do.

Here, D is a local window of 7 x 7 and represents the local area around (u, v). N is the number of pixels in D. θ (i, j) and θ (u, v) are the local ridge directions of pixels (i, j) and (u, v), respectively.

In the step of comparing the consistency level and the consistency threshold, the consistency level is compared with a preset consistency threshold value Tc (S323). If the consistency level is less than the consistency threshold value Tc, the current consistency level is compared. The resolution of the image is lowered, and the process returns to the local direction field estimation step ( S321 ). That is, if the coherence level is below the coherence threshold Tc, the local orientation of this region is re-estimated at a lower image resolution level until the coherence exceeds Tc.

In the step of comparing the consistency level and the consistency threshold, as a result, a direction field exceeding the consistency threshold is obtained.

In the adaptive filtering step (S324), if the consistency level of the direction field is greater than or equal to the consistency threshold in the step of comparing the consistency level and the consistency threshold, the filtered image (that is, the filtered image in the step of applying the Gabor filter bank) image), the following two adaptive filters are applied, namely, two masks, h _t (u,v;i,j) and h _b (u,v;i,j).

where θ (u, v) represents the local ridge direction of the pixel (u, v). These two masks, i.e. h _t (u,v;i,j) and h _b (u,v;i,j), are the local maximum gray along the default direction of the local ridge direction (i.e. the normal direction). Adaptively emphasizes (enhances) the level value. The filtered image is first _{combined with two masks: h t} (u,v;i,j) and h _b (u,v;i,j).

In the step of comparing the gray level value and the ridge threshold value, the gray level value of the pixels (u, v) of the calculated image, that is, the convolved image, is set to a preset ridge threshold. Comparing with the value (T _ridge ) ( S325 ), if the gray level value of the pixel (u, v) of the calculated image, that is, the convolved image, is a preset ridge line If it is greater than the threshold value (T _ridge ), the pixel (u, v) is first designated as a ridge label (S326).

Adjusting the mask width to the width of the local ridges allows this algorithm to efficiently find ridges in the fingerprint.

Because the input fingerprint image has noise, wrinkles, and specks, etc., the resulting ridge map of the filtered image contains a large number of non-ridge pixels, often labeled as ridge pixels. To remove these non-ridged pixels, a fake ridge removal step is required.

In the step of comparing the area of each component of the ridge map with the minimum threshold value, the area of each connected component (each major element) appearing on the ridge map is calculated (S327), and the calculated area is compared with the minimum threshold value (T _min ) Compare (S328), and if it _{is smaller than the minimum threshold value (T min} ), it is designated as a background label (S329), otherwise, the connected component is divided into a set of short line segments (S330).

In the step of determining whether a line segment is located between a pair of parallel ridges, it is determined whether each short line segment is between a pair of narrow parallel ridges (S331), and if so, label it with an actual ridge line (S332), otherwise the background It is designated as a label (S332).

Here, the step of comparing the area of each component of the ridge map with the minimum threshold value and the step of determining whether a line segment is positioned between a pair of parallel ridges may be referred to as a fake ridge removal step. Most of the fake ridges are removed by performing the fake ridge removal step in the ridge map.

The coarse level ridge map and the unrecoverable area mask generation step (S350) is performed by the quality mask generation unit 60, after the ridge extraction step (S320), votes in the extracted ridge map of the filtered image (S320) voting) algorithm is used to generate a coarse level ridge map and a non-recoverable area mask.

That is, after obtaining the ridge map of each filtered image in the ridge extraction step S320, a coarse-level ridge map of the input fingerprint image and a mask of an unrecoverable region are generated, and the coarse-level ridge map is generated. The ridge map is used to estimate reliable direction fields. The generated coarse level ridge map should roughly reflect the direction of the local ridge/valley structure of the input fingerprint image. However, this coarse level ridge map need not be very accurate in terms of local ridge structure.

In the enhancement algorithm of the present invention, a coarse level ridge map and an unrecoverable region mask are generated from the ridge map of the filtered image using a voting algorithm.

In the step of dividing into blocks, each ridge map in the filtered image is divided into blocks of size W x W (eg 8 x 8).

Comparing the number of ridge pixels with a threshold for the number of ridge pixels, comparing the number of ridge pixels around a block (that is, within a block) to see if it is greater than or equal to a preset ridge pixel number threshold, and if so, if there are enough ridge pixels around the block If it appears (i.e. if the block has enough ridge pixels), we label each block as foreground (value 1), otherwise we label each block as background (i.e. background) with label (value 0).

After the step of comparing the number of ridge pixels with the ridge reference number of pixels, the binary block map with a pixel value of 1 indicates that there is a ridge, and 0 indicates that there is no ridge in each ridge map of the filtered image. That is, all connected components (8 connections) are deleted from the binary block map having an area below the threshold (eg, 16). In this way, all eight filtered images are checked for each block.

This is a coarse level ridge map calculation step, and a coarse level ridge map is calculated according to the following rule.

First, if only one of the 8 binary block maps of pixel (x, y) has value 1 and this pixel belongs to a connected component of size K (provided that K is _{greater than the block threshold (T block} ), then K> T _block ), the corresponding block pixel value in the coarse level ridge map is duplicated in the associated ridge map. The pixel value of that recoverable region mask is set to value 0 to indicate that this block can be recovered.

Second, if two or more binary block maps at a pixel (x, y) have vlaue 1 and the connected local ridge directions are not orthogonal to each other, the average value of the ridge map values associated with the pixel values of the corresponding blocks in the coarse level ridge map is considered The pixel value of the corresponding recoverable region mask is set to value 0, indicating that this block can be recovered.

Third, if two or more binary block maps have a value of 1 at a pixel (x, y), the associated local ridge directions can be orthogonal to each other, and the connected component has a pixel whose size is larger than the _{block threshold (T block ).} If there is only one (i.e., only one pixel with a value of 1, in _{a connected component of size greater than the block threshold (T block} )), then in the coarse level ridge map, the pixel value of that block has the largest connected component. It is copied to the ridge map associated with the component and the pixel value of the corresponding recoverable area mask is set to value 0 to indicate that it is a recoverable block.

Fourth, if the above conditions are not met, the block is assigned the label 1 indicating that it is not recoverable.

If this algorithm (that is, the coarse level ridge map and the unrecoverable region mask generation step S350) with such a rule is applied to the ridge map set of the filtered image, the coarse level ridge map and a non-recoverable area mask is created.

In the direction field estimation step ( S360 ), the direction estimation algorithm is applied to the coarse level ridge map and the coarse level ridge map generated in the unrecoverable region mask generation step ( S350 ), so that the local get direction

The coarse level ridge map generated from the ridge map of the filtered image maintains the local direction information of the ridge/valley structure of the input fingerprint image. Through the coarse level ridge map and unrecoverable area mask generation step S350, the direction field of the specified input fingerprint image is ignored in the unrecoverable area and stably in the coarse level ridge map can be estimated

f _i (x, y) (where i = 0,1,2,3,4,5,6,7) is the gray (x, y) in the pixel (x, y) of the filtered image corresponding to the direction ( θ _{i )} gray) represents the level value, and represents θ _i = i × 22.5°. After estimating the direction field, the gray level value of the pixel (x, y) of the enhanced image can be obtained according to the following formula:

θ(x, y) represents the local direction field value of pixel (x, y).

As a post-processing step (S370), a fake prime is displayed using the following objective function (E).

where θ _i and θ _j are the angles where the minutiae i and j are dependent on a fixed reference axis, B is the average brightness of the joining line, and D is the distance between them. A value of E (i.e., a target value) is calculated for each pair of primes that are closer (smaller) than the distance threshold Dth. If the value of E is greater than the target threshold (Eth), both features (minutiae) are deleted and replaced with a line connecting the original features (minutiae). For this connecting line, information on the correct connection and number of ridges can be easily obtained.

In general, algorithms alone cannot always perform correctly in all situations. Also, in practice, enhancement algorithms may introduce their own artifacts. Therefore, a post-processing technique is needed to reduce the extraction of unnecessary decimal points due to this effect.

The key to designing these post-processing strategies is to anticipate these situations where the enhancement algorithm will not be effective. Post-processing techniques were employed to detect and overcome each of these difficult individual situations. The present invention applies a post-treatment technique to treat postpartum cracks that cannot be treated by an enhancement algorithm.

Cracks in the fingerprint image can generally be characterized as narrow bright areas running transversely in the direction of the dominant ridge. One of the inferences for detecting minutiae due to these cracks is based on the observation that the fractional terms are semi-aligned and the region between them is brighter than the average brightness of the foreground region. The present invention obtains the information necessary to identify these primes from the input fingerprint image and the set of detected primes in the enhanced image. Then, a fake prime is displayed using the objective function described above.

8 is a result of applying the improved algorithm of the present invention to a fingerprint image with poor quality.

Fig. 8(a) is an input image, Fig. 8(b) is a coarse level ridge map, Fig. 8(c) is an unrecoverable area mask composed of white pixels, (d) is an expected direction field, (e) of FIG. 8 is an enhanced image, and (f) of FIG. 8 is a minutiae extracted from an enhanced image superimposed on an input image.

The present invention is not limited by what has been described above and illustrated in the drawings, and it is a matter of course that those skilled in the art can make many more modifications and variations within the scope of the claims described below.

10: fingerprint quality improvement module 20: Gabor filter bank unit
50: ridge extraction unit 60: quality mask generation unit
70: direction field estimation unit 80: post-processing unit

Claims (9)

Gabor filter bank unit, using a Gabor filter bank, filtering the input fingerprint image, generating a filtered fingerprint image, Gabor filter bank application step;
A ridge extraction step of generating a ridge map by extracting a ridge line from each filtered fingerprint image output in the Gabor filter bank application step;
The quality mask generator applies a voting algorithm to the ridge map generated in the ridge extraction step to generate a coarse-level ridge map and an unrecoverable area mask, a coarse-level ridge map and recoverable generating a mask of an absent region;
The direction field estimator may include: a direction field estimation step of obtaining a local direction of each pixel from the coarse level ridge map and the coarse level ridge map generated in the unrecoverable region mask generating step;
Fingerprint enhancement method comprising a. According to claim 1, wherein the ridge extraction step,
a local direction field estimation step in which the ridge extraction unit obtains a local direction field from the fingerprint image filtered in the Gabor filter bank application step;
_{The ridge extraction unit obtains the consistency level (C 0} (u,v)) of the direction field obtained in the local direction field estimation step in the local area of the pixel (u, v), but the consistency level is a preset consistency threshold calculating the consistency level of the direction field, obtaining a local direction field exceeding the value Tc;
an adaptive filtering step of applying, by the ridge line extraction unit, two masks composed of an adaptive filter to the fingerprint image filtered in the Gabor filter bank application step after the coherence level calculation step of the direction field;
When the gray level value of the pixels (u, v) of the image to which the two masks of the adaptive filtering step are applied is greater than the _{preset ridge threshold value (T ridge ),} Comparing the gray level value and the ridge line threshold value in which the area is calculated by the ridge line extraction unit;
The ridge line extraction unit determines whether the area calculated in the step of comparing the gray level value and the ridge threshold value _{is smaller than the minimum threshold value (T min} ). If it is small, it is designated as a background label. Comparing the area of each component of the ridge map and the minimum threshold value by dividing the ? into a set of line segments;
The ridge extractor determines whether each line segment divided in the area of each component of the ridge map and the minimum threshold value comparison step is between a pair of parallel ridges, and if each line segment is between a pair of parallel ridges, it is an actual ridge line. determining whether a line segment is located between a pair of parallel ridges by designating a label, and designating a label as a background label if each line segment is not between the pair of parallel ridges;
Fingerprint enhancement method comprising a. According to claim 1,
The specific objective function (E) is

(where θ _i and θ _j are the angles at which the features i and j depend on the fixed reference axis, B is the average brightness of the line, and D is the distance between the features i and j)
a post-processing step in which the post-processing unit displays a fake prime number by using ;
Fingerprint enhancement method further comprising a. According to claim 3, Gabor filter bank step,
After the Gabor filter bank unit performs FFT on the input fingerprint image to obtain a frequency image, a corresponding Gabor filter with predetermined radial and directional frequencies is applied to the frequency image, and an inverse FFT is performed on the frequency image to which the Gabor filter is applied. A fingerprint enhancement method comprising obtaining a filtered image. 5. The method of claim 4,
The Gabor filter has a center frequency of 60 cycles/width (height), a radial bandwidth of 2.5 octaves, and _{8 values of the center direction θ 0} are 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees. , 135 degrees and 157.5 degrees, and the direction bandwidth is 35 degrees.
3. The method of claim 2,
In the adaptive filtering step, the two masks (h _t (u,v;i,j) and h _b (u,v;i,j)) composed of the adaptive filter are

(however, θ (u, v) represents the direction of the local ridge line of the pixel (u, v))
Fingerprint enhancement method, characterized in that. 3. The method of claim 2,
In the coles level ridge map and irrecoverable area mask generation step, the quality mask generation unit,
Each ridge map of the filtered image output from the ridge extraction unit is divided into blocks of size 8 x 8, and if the number of ridge pixels around each block is greater than or equal to the preset ridge pixel count threshold, each block is converted into a foreground ( label with value 1), and if the number of ridge pixels is less than the threshold number of ridge pixels, each block is designated as a background (background) label (value 0),
Fingerprint enhancement, characterized in that a coarse level ridge map and an unrecoverable area mask are generated by applying a coarse level ridge map according to a preset coarse level ridge map calculation rule Way. 8. The method of claim 7,
The ridge map calculation rules for the coarse level are
If only one of 8 binary block maps of pixel (x, y) has value 1 and this pixel belongs to a connected component of size K, when K is _{greater than the block threshold (T block} ), the coarse level a first rule, wherein a corresponding block pixel value in the ridge map is duplicated in an associated ridge map, and a pixel value of the corresponding recoverable region mask is set to value 0 to indicate that the block can be recovered;
If two or more binary block maps at a pixel (x, y) have vlaue 1 and the connected local ridge directions are not orthogonal to each other, the pixel value of the corresponding block in the coarse level ridge map is regarded as the average value of the associated ridge map values. and the second rule, wherein the pixel value of the corresponding recoverable region mask is set to value 0 to indicate that the block can be recovered;
If two or more binary block maps at a pixel (x, y) have a value of 1, the associated local ridge directions can be orthogonal to each other, provided that the connected component has only one pixel whose _{size is greater than the block threshold (T block ).} , the third rule, in which the pixel value of the block in the coarse level ridge map is duplicated in the ridge map associated with the largest connected component and the pixel value of the corresponding recoverable area mask is set to value 0 to indicate that it is a recoverable block. and,
A fingerprint enhancement method comprising a fourth rule in which a value of 1 is assigned as a label indicating that the block cannot be recovered if the first to third rules are not met. 3. The method of claim 2,
In the direction field estimation step, the direction field estimation unit,
After finding the local direction of each pixel in the coarse level ridge map, the gray level value (g(x,y)) of the pixel (x, y) of the enhanced image is calculated

(where θ(x, y) represents the local direction field value of pixel (x, y))
A fingerprint enhancement method, characterized in that obtained by.

Images