Feature Extraction And Matching Of Finger Vein Image

In this paper we analyze many kinds of biometric identification technologies at the beginning, and compare the difference of them. Although many biometric identification technologies like finger print have applied broadly now, they can be copied and forged easily because they are exposed outside the human body, for avoiding this defect, persons like Naoto Miura proposed the biometric identification system using patterns of veins within fingers. Finger-veins within fingers cannot be copied and forged, and have relative uniqueness, and can be used with finger print or palm print too. Based on the repeated line tracking method and template matching method advanced by Naoto Miura ,this paper achieves and improves the algorithms ,and proposes the idea of moment invariant method to match finger-vein .The method of repeated line tracking is : firstly, capturing finger-vein images with infrared light ,then selecting some pixels in the image randomly ,beginning at these pixels ,tracking local dark lines pixel by pixel ,when a dark line cannot be detected ,a new tracking process begins from another new position ,all the dark lines in the image can be tracked by repeatedly extracting such local line tracking operations. finally, the loci of the lines overlap and the pattern of finger veins is on turned statistically ,as the part of the dark lines are tracked again and again in the repeated operations they are increasingly emphasized. When from current tracking pixel of line next one, the tracking direction is selected randomly, there are five possibilities: three pixels in the left, three pixels in the right, three pixels of the up ,three pixels of the down, and eight neighboring pixels.In this paper, if we selected tracking direction randomly like the method of Naoto Miura, there will appear the phenomenon of not instability finger-vein images. And this cannot be changed through enlarging the count of random pixels, so in the paper, when fixing on the next pixel, we select the eight neighboring pixels as the possible next pixel, so the instability phenomenon is avoided.In practice, there must be some preprocesses before finger-vein extraction. We adopt algorithms of removing background and Gauss filter to delete noises in this paper. After vein extraction, there is still noise in the image, in this we adopt some postprocesses to delete noises. Firstly, we delete the isolated pixels and connect two break pixels, then we carry on average value filter processing. For carrying on moment invariant, we adopt the thinning algorithm of image, after the thinning process, we delete noises again, and we obtain the finger-vein of feature extraction at last.Template matching algorithm is: before matching formally, we identify the ambiguous regions firstly and the slight misalignments between vein patterns .Robust template-matching is thereby achieved. Then we fix on a rectangle region, computing the mismatch value as two finger-vein images aim at every pixel in the region ,at last ,selecting the smallest value as the final mismatch value .If the value less than the threshold we initialized beforehand, we consider they matched, else, we consider they don't matched. The veracity of template matching is high, but when searching millions of images, the time it uses is too long .because of this ,this paper proposed the moment invariant algorithm to match the finger-vein images.The definition of moment invariant is: for the duality limitary function f ( x , y ),the( j + k) rank moment is Moment invariant is a geometry characteristic related to the planar region .Most characteristics like size , position, direction, and pattern and so on ,are related to the moment ;The one rank moment is related to the pattern ,the two rank moment is related to the enlarging degree of the average value of curve circumfuses lines; the three moment is the measure of symmetry of average value.In this paper, in the thinning finger-vein image, take the gray value of every pixel as the duality limitary function f ( x , y ),and match finger-vein images using Hu moment invariant as follows: Then distinguish result using distance measurement formula: Fai[ i ] express the seven moment invariant of the first thinned image, Fai1 [ i ]express the seven moment invariant of the second thinned image. We fixed on the threshold as 2.5, if Num is less than 2.5, we think the two images as the same finger, and the result is that they matched, whereas we think they aren't the same finger, and they didn't match.The algorithm of this paper has been tested on the finger-vein images captured by Changchun Hongda Company. The testing results demonstrates that our moment invariant scheme satisfy the need of speed , the time it spend is less than 1ms, but the accuracy still need to be tested through more high quality finger-vein images , and can be improved much more. In addition, we can improve the algorithm from the type of moment invariant, considering some other moment invariant methods, we can improve the measurement distance to better the algorithm. In view of the improvement of matching methods, we should improve the preprocesses of finger-vein images correspondingly.