KR20000034228A - Method of recognition finger print using direction information of wavelet - Google Patents

Abstract

PURPOSE: A method of recognition finger print using a direction information of a wavelet is provided so that a real time finger print recognition can be implemented by reducing an amount of computation on the finger print containing complex patterns. CONSTITUTION: A method of recognition finger print using a direction information of a wavelet includes several steps. At first, if a finger print image is inputted (301), a wavelet conversion step is executed (302). Thereafter, a step is to seek information on coherence and local directions (303) and to discriminate whether there exists the registered finger print (304). As the result, if the registered finger print does not exists, a process is finished after recognizing the finger print as a different finger print (305). If the registered finger print exists, a step is to call the registered finger print visualized (306). Thereafter, a step is to compute a move information (a region of LL) (307) and discriminate whether a coefficient of correlation is larger than the critical value of the region of LL (308). If the coefficient of correlation is larger than the critical value, a step is to recognize the finger print as a same finger print (311).

Description

Fingerprint Recognition Method Using Directional Information

The present invention uses the directional information of the ridges to recognize the fingerprint in real time, in particular a fingerprint recognition method. Smoothing and binarization are applied because coherence and dominant local orientation are applied in the wavelet transform region. In order to recognize the fingerprint in real time by omitting preprocessing such as thinning and restoration.

Hereinafter, a fingerprint recognition method according to the prior art will be described.

Conventionally, when fingerprint data is input, a preprocessing process is performed.

The preprocessing process is a preparation process for extracting features from the input image.

In this preprocessing, there are stages such as binarization, thinning, and correction.

After such a preprocessing process, the feature extraction process of extracting the feature of the fingerprint is performed, which becomes a factor of recognition.

After that, it is a process of determining a feature group with common parts of the recognition factor through a learning process.

And finally, as a comparison process, a recognition factor between the referenced fingerprints and the input fingerprints is compared.

The fingerprint input through the above processes is recognized. There is a method of extracting the directional component of the ridge in the form of pixels through the binarization during the preprocessing.

This method determines the color and brightness levels of the data by 0.255 black and white, and uses the direction information of the portion that appears in black as a factor.

Meanwhile, instead of the binarization process, there is also a method of extracting directional components of the ridge using the minimum brightness difference method.

In addition, in order to implement a recognition factor better, there is a method of extracting features through a thinning process as well as a binarization process.

Factors extracted through the method (objects of recognition), as described above, create a feature group consisting of the same factor for each fingerprint through the learning process.

The comparison process is a comparison process between input data that has not been learned and learned reference data.

Among the methods of the comparison process, there is a method of matching by syntactic method and a method of matching using a graph.

In addition, there is a method of combining by a structural matching method, a method of combining a statistical method and a structural matching method.

However, the conventional fingerprint recognition method as described above has the following problems.

In order to extract better and more reliable features, the pretreatment process is very important.

This is the biggest obstacle to the real-time implementation of the recognition, and the binarization during the preprocessing process may have a large error in the extracted direction components when the binarization process is performed on the data of poor image quality.

In addition, if the minimum brightness difference method requires a considerable amount of overlap calculation, the process of thinning requires a great deal of processing time.

Therefore, it was difficult to recognize the fingerprint in real time.

The present invention has been made to solve the above problems, and an object thereof is to provide a fingerprint recognition method using direction information of a wavelet capable of recognizing a fingerprint in real time.

1A to 1B are views illustrating an original fingerprint image and a wavelet transformed fingerprint image for explaining the present invention.

2A to 2C are fingerprint images when Gaussian is applied and fingerprint images when Coherence is applied and fingerprint images when Dominant Local Orientation (DLO) map is applied.

3 is a flowchart illustrating a fingerprint recognition method using wavelet transform of the present invention.

Fingerprint recognition method using the direction information of the wavelet of the present invention for achieving the above object is to extract the direction component of the ridge using the wavelet transform when the fingerprint image is input, and the wavelet transformed fingerprint image through the gradient Gaussian Obtaining a direction vector, extracting coherence data and main local direction information using the same, and defining a fingerprint image as a small block, and then applying the mean and the variance of each block to the rows and columns to obtain a variance value. Setting the largest block as a reference block, obtaining a maximum correlation coefficient between the image of the fingerprint registered and the input fingerprint based on the reference block, and defining the image of the input fingerprint for each zone and sequentially Comparing the thresholds of the respective zones with the thresholds and the correlation coefficients of the final zones. Determining that the input fingerprint matches the registered fingerprint if the number is greater than the threshold of the last zone; otherwise determining that the fingerprint does not match.

Hereinafter, a fingerprint recognition method using the direction information of the wavelet of the present invention will be described with reference to the accompanying drawings.

First, in the present invention, there is no need for binarization of an image and almost no overlapping calculation is required.

The signal can be localized simultaneously in the time (space) -frequency domain, and the wavelet transform is used to extract the characteristics of the signal using the properties of the bandpass filter and to suppress the noise under the influence of the lowpass filter. The fragrance component of the ridge was extracted.

At this time, the extracted aromatic components were localized by multi-resolution, and then characterized by using a gradient Gaussian.

That is, the present invention intends to obtain features using firstly the characteristics of wavelet transform and secondly using new measurer coherence and main local direction so that a large part can be omitted in the preprocessing process.

The first and second methods are used to extract features using a fingerprint processing algorithm having a small number of complex patterns, and the recognition algorithm in the multi-resolution domain brings about a reduction in the computation amount.

This will be described in more detail as follows.

First, the fingerprint recognition method using the direction information of the wavelet according to the present invention is a step of using a wavelet transform as a first step.

FIG. 1A is an original fingerprint image, and FIG. 1B is a wavelet transformed fingerprint image.

The first step, the wavelet transform, is designed using a symmetric wavelet. When the original fingerprint image is subjected to the wavelet transform, the wavelet transform appears as shown in FIG. 2B.

Next, as a second step, there is a direction vector feature extraction step by a gradient Gaussian.

This step executes the gradient using the IDD algorithm on the wavelet transformed image and then executes coherence again.

The main local direction is obtained using the data of coherence.

2A to 2C illustrate the second step, FIG. 2A is a fingerprint image when Gaussian is applied and FIG. 2B is a fingerprint image when Coherence is applied. 2C is a fingerprint image when a Dominant Local Orientation (DLO) map is applied.

2A to 2C illustrate the second step, FIG. 2A is a fingerprint image when Gaussian is applied and FIG. 2B is a fingerprint image when Coherence is applied. 3C is a fingerprint image when a Dominant Local Orientation (DLO) map is applied.

Subsequently, the third step relates to reference block setting.

This step is necessary to recognize the registered fingerprint and the input fingerprint at the same location.

In the comparison of the base images, the designation of the reference position is very important. Therefore, the method of determining the reference point is an important factor that directly affects the performance of fingerprint recognition.

Use mean and variance to establish these reference points.

First, the initial reference point in the effective area of 9x9 blocks is found around the center point of the LL area whose total size is 16x16 blocks.

Using the phenomenon that the inflection occurs more severely toward the center of the reference, the initial reference point is defined as the point where the maximum variance values of the columns and rows meet.

The reference point thus determined approximately coincides with the center point of the fingerprint. However, in the relationship between the enrolled fingerprint and the input fingerprint, some errors may occur due to various physical factors.

Therefore, a finer reference point, that is, the setting of the final reference point is required.

To this end, in the present invention, the final reference point is determined using cross-correlation.

Specifically, the final reference point is determined by using the cross-correlation of all points between the enrolled fingerprint and the input fingerprint existing in the 4x4 block with respect to the initial reference point determined above.

Next, the fourth step is to distinguish between the enrolled fingerprint and the input fingerprint.

This step is the final step of fingerprint recognition and proceeds in the following order.

First, the center points of the LL, LH, and HL regions are mapped from the input fingerprint image to the final reference point obtained in the above-described three steps.

After mapping the LL region of the registered fingerprint and the LL region of the input fingerprint, the correlation coefficient values are obtained.

If the obtained correlation coefficient values are smaller than the predetermined threshold, one point is given. If the given score is greater than the threshold for each region, the next stage of mapping of the LH region is started, otherwise it is recognized as a fingerprint of another person.

The same goes for the LH area and the HL area.

Therefore, the final recognition method recognizes the same fingerprint if it is larger than the threshold for each region in all regions.

The process from step 1 to step 4 described above is shown in the flowchart of FIG.

As shown in FIG. 3, when a fingerprint image is input (301), wavelet transform is performed (302).

Thereafter, the coherence and local direction information is found (303), and it is determined whether a registration fingerprint is present (304). As a result, if the fingerprint is not present, it is recognized as another fingerprint (305), and if the fingerprint is present, the registered fingerprint is called (306).

Then, the movement information is calculated (LL area) (307), and it is determined whether the correlation coefficient is greater than the threshold of the LL area (308).

As a result, if the correlation coefficient is not large, it is again determined whether there is a registration fingerprint (304). If the correlation coefficient is larger than the threshold for each region, this time, it is determined whether the correlation coefficient and the threshold for each region are large or small (309). . That is, if the correlation coefficient is not greater than the threshold of the LH region, it is determined whether the registration fingerprint exists (304). If the correlation coefficient is greater than the threshold, this time, the threshold and the correlation coefficient of the HL region are determined (310), and the correlation coefficient is HL. If it is not greater than the threshold of the area, it is again determined whether there is a registration fingerprint (304). If the correlation coefficient is greater than the threshold, it is recognized as the same fingerprint (311), and the fingerprint recognition process is terminated.

As described above, the fingerprint recognition method using the direction information of the wavelet of the present invention has the following effects.

Fingerprint recognition can be implemented in real time by extracting features using a small number of complex patterns of fingerprints and reducing the amount of computation.

In addition, there is an effect that can find in real time not only the fingerprint recognition but also the same pattern, such as letters, license plates, fabric defects, inconsistent patterns.

And it can be used to speed up the fingerprint automatic system for home security, access control of confidential area, police crime investigation, access control of computer system, etc. as hardware.

Claims (1)

Extracting a direction component of the ridge using a wavelet transform when a fingerprint image is input, Obtaining a direction vector from the wavelet transformed fingerprint image through a gradient Gaussian; Extracting coherence data and main local direction information using the same, Defining a fingerprint image as a small block, and then applying the average and the variance of each block to the rows and columns to set the block having the largest variance as the reference block, Obtaining a maximum correlation coefficient between an image of a fingerprint registered based on a reference block and an image of an input fingerprint; Defining an image of an input fingerprint for each zone and sequentially comparing the threshold and the correlation coefficient for each zone; Comparing the threshold value of the last zone with the correlation coefficient and determining that the input fingerprint matches the registered fingerprint if the correlation coefficient is greater than the threshold value of the final zone; Fingerprint Recognition Method using Wavelet Transform.