A method for enhancing the limitations of fixed wave ridge orientation beach, like reference L.Hong, the methods section of the present study using Gabor way multiplexer fingerprint image filtering, this method is critical to the fingerprint ridge and a groove direction frequency estimates string line. So using a frequency estimation method proposed by the authors fingerprint ridge has been improved to enhance its performance. The enhancement method to filter out the structure of my voice on the original fingerprint image and voice sound statistics.
First, the principle of image enhancement characteristics and functions of Gabor
In the fingerprint image, alternating parallel ridges and valleys appears to produce good orientation and frequency information. Based on this property. Direction and having a reinforcing effect by using a frequency selection characteristic of the band pass filter can be obtained Da good wave ripple.
Expression applies image processing Gabor filter in the spatial domain.
u, v is a variable filter Jia bits; 8x, 8y, collusion with a frequency and direction of the pass band, the filter direction angle is 0, f is the filter center frequency. In the spatial domain, Gabon Beach wave is constituted by the product of the two-dimensional Gaussian function and a sine function, the Fourier sine function is converted into the sub impact should noon, the Fourier transform of the Gaussian function remains Gaussian function, so frequency wave train should noon Gaborr for the impact convolution function and Gaussian function. FIG shows the empty city and town image frequency ripple Gabor filter.
Frequency response from the image can be seen, Gabor filter pass band which has good properties.
Thus, according to the corresponding local ridge line direction and a ridge configured Gabor ripple frequency, a partial fingerprint image and ripple, just the filter passband the local ridge orientation and frequency distributions overlap, can enhance the local ridge while the line direction and frequency information Zi out other voice sound signals, thereby achieving fingerprint image enhancement.
Second, the fingerprint ridge direction calculation
Fingerprint pattern reflects the nature of the fingerprint image texture. In this paper, the gradient operator obtaining the proposed method Rao direction in FIG. A pixel (i, j) disposed at the fingerprint image gray gradient [ax (i, j), ay (i, j)], where the grayscale gradient vector may be calculated Sobel gradient operator.
Reflects the direction of shades of gray at the fastest pixel density change direction, leisure partial fingerprint ridge is disposed over the planar sine wave, the direction of the gradient in the vertical direction and the pixels of the fingerprint ridge, fingerprint thus obtained block diagram showing the basic direction to be able to show a fingerprint ridge, but in the voice sound more serious impact areas will appear in the wrong direction. Low pass smoothing filter fingerprint entire sub block directions can solve this problem. Further If the two differ Wan fingerprint ridge in direction, it is considered that two stripe-line in the same direction, i.e. the direction of the gradient distribution in [0, @]. For this purpose, the gradient vector argument twice expanded so as to extend [0,2x], and ultimately the calculated directional angle is further reduced 1/2.
Directional FIG specific steps as described above are as follows:
(1) the image into the image of the sub-blocks wxw:
(2) calculate the gradient G of each pixel within each sub-block (i, j) values ax (i, j) and ay (i, j).
(3) direction of the image sub-block is calculated.
(4) the direction of the sub-blocks into a continuous low-pass filtering to the bowl field.
Obtained as described above fingerprint pattern, the point direction is a direction of each pixel belongs to the ridge, all of the pixels of the block orientation directions within a sub-block is approximately unity.
Third, the trouble rate calculating Fingerprint
In the local area range of a fingerprint, or if there is no singular minutiae points, ridges and valleys along the gray line it can be seen as a sine wave distribution perpendicular to the ridge direction, as shown in FIG. Sine wave pattern and the direction of the window of the X-coordinate direction along a line in the direction of the window (OrientedWindow) in gray projection (x-signature) is formed with the same frequency. So it can be estimated by averaging the local ridge frequency sine wave signal frequency is calculated projection.
Represents the normalized image with G, represents a fingerprint opening direction of the image, the fingerprint ridge frequency is calculated according to the following steps:
(1) the image G image is divided into m sub-blocks of WxW:
(2) a center point (i, j) as the center of each sub-block, the window size lxw construct direction by the coordinate system shown in FIG;
(3) Calculate the gray value ridge projected on the x-coordinate direction of each sub-block window.
(4) located at x [k] bits of all the peaks, calculate the distance between all two adjacent peaks, and then determined.
The average distance D (i, j), the current sub-block ridge frequency f (i, j) of D (i, j) effect inverted.
However, in some regions or voice sound feature-point area, the projection can not form good sinusoidal signal, thus determining method ridge frequency is not accurate enough to calculate the frequency analysis method described herein use frequency projection signals Yi by computing the average peak distance, reducing according to the gradation frequency projection signals directly calculate the error caused by the string.
Sinusoidal signal frequency value can be obtained by analyzing the Fourier transform thereof bit spectral peaks. However, the projection area of signal noise or the feature point is not the normal reaction zone ridge frequency, and therefore can not simply be determined by its frequency spectrum string corresponding to the major peak, which is used for this purpose high-order spectrum analysis method to calculate the fundamental frequency signal .
Fourth, the fingerprint image enhancement algorithm Gabor sent Boshu
The gradation image referred to as the normalized G, after the enhanced image is E, using discrete convolution values obtained for each pixel point enhancement.
Wherein the formula is obtained by the Gabor filter mask, mask hole size of the filter. Each ridge direction corresponding to the image taken beach wave for the image block, the pre-image of each corresponding chaotic wave are taken for the image block ridge frequency. Passband size provided should also have a certain influence on the reinforcing effect of the fingerprint image, when the pass band is large, a small inhibitory effect on the noise, the reinforcing effect is not obvious but can tolerate a certain degree of error direction and the frequency of the cry; when the pass band is small , direction and frequency selective filter is strong, has a good effect on the fault line and the connection is broken even ridge separating dry, but more sensitive to errors provided cry Beach wave direction and frequency. After a number of experiments found that when the size of the beach wave 11, the wave effect can be obtained very good way. The algorithm is as follows:
(1) the size of the fingerprint image is divided into sub-blocks and do not re-registration of the WxW;
(2) using an algorithm described above were calculated for each sub-block image ridge orientation and ridge frequency;
(3) using the formula and the corresponding parameters for each subimage corresponding enhanced templates were constructed;
(4) for all sub-blocks of the image, with its ripple calculation formula, a result of image enhancement.
FIG show the experimental results of the algorithm, the original fingerprint image in FIG. 1, FIG. 2 is a section using a method for enhancing the resulting image.
Seen from the image, this section fingerprint enhancement algorithm has very good overall enhanced fingerprint dry cracking and even effect, but at a certain reinforcing effect singular region is not particularly desirable.
This section examines the fingerprint image to the beach wave tomb Gabor function enhancement algorithms. Ridge orientation calculating average gradient method, pour rate is calculated by the ridge structure local gray ridge projection model. To minimize these errors, the projection signal spectrum analysis that more accurately calculate the frequency sequence. Since the ridge orientation calculated value is continuous, the method is superior in accuracy to the fixed direction filtering method, and achieved good reinforcing effect. However, the operational process requires pre individually calculating direction and rate, the operation cost is relatively large, suitable for very high speed is not the case, and the frequency accuracy of the calculation depending on the accuracy of the calculated direction crude.
Transfer: https: //blog.csdn.net/ankertqf/article/details/81238102