Research On Algorithm Of Vein Recognition

Biometric technology, also named Biomensuration technology, use human being's physical character ( such as finger print, iris, facial features and so on ) and behavior character ( such as signature, voice and so on ) to recognize person's identity automatically by using the method of image processing and pattern recognition. The physical characters of human being include voice, finger print, palm print, eyeground, appearance, DNA, signature and so on. The technical core of biometric lies in how to obtain these characters, then transfer them into digital information and store in the computer, recognize person's identity by using reliable matching algorithm. Vein is one of the most strong and permanence physical characters. Vein recognition has many advantages such as high recognition ratio, difficult to forge, not to be influenced by outer condition and only can recognize on vivid. So it becomes a recognition method easy to be accepted.The construction of geometric image representation is a very active research area where many beautiful and innovative ideas have been tested. Recently many way about Multiscale Geometric analysis were produced, including Ridgelet transform, Monoscale ridgelet transform, Curvelet transform, Bandelet transform, Contourlets transform and so on. Bandelet transformation is an image representation method based on edges. It can track the geometric regularity of the image self-adapted. This method define a kind of geometric vector lines to token the local regularity of the image, then 2-D partition the support area. When the partition is thin enough, there is only one edge of the image in every partition section. Geometric regularity need not to be defined for the partition sections which have no edge. While for the partitionsections with edge, the geometric regular direction is just the tangent direction of the edge. According to the global best restriction of the local geometric regularity, vector lines in the partition sections were calculated, then along the vector lines to Bandeletzation the wave in this section and create the Bandelet base.We use the advantage of the Bandelet transformation to put forward a vein recognition algorithm based on Bandelet transformation. The first step of the algorithm is to Bandelet decompose a 256×256 unitary image, then make tow times wave transformation for the low frequency part. The subimage in high frequency part include the most noise, they are not fit for feature extraction. So we use seven subimages with shadow in the following figure to do feature exaction. In order to get more vein feature, we move a 4×4 template in the three shadows in the top left corner to get 48 feature points. And then move a 8×8 template in the two bigger shadows in the top left corner to get 32 feature points. At last we move a 16×16template in the two biggest shadows to get 32 feature points. There are 112 feature points altogether. For each point we calculate the mean and difference variance to form a 2-D feature vector. In the recognition phase we calculate the coefficient of two sample's feature vectors to get their similarity degree.2-D Gabor wavetransformation is an important tool of signal analysis and processing in time-frequency field. Its transformation coefficient has fine visual speciality and biology background. It is widely used in the fields of image processing and pattern recognition. On the basis of Daugman's iris recognition algorithm based on 2DGabor filter and Zhang's palm print recognition algorithm based on 2DGabor filter, we present a vein of finger recognition algorithm based on plural 2-DGabor transformation and a new vein encode method combined the phasic information and orientation information——Vein Phasic Orientation Code(VPOC). We use annular 2D filter to do feature exaction. 4 filters with same frequency and variance and different orientation are selected to exact VPOC. The orientation are 0 0, 450 , 90 0 and 1350 respectively, expressed as Gi ,where i = 0,1,2,3. By Gi , the vein image is filtered, the filtered image is defined as And the orientation of point ( x ,y ) can get by the following formula: The phasic code of vein can get by the following 2 formulae: We can define VPOC of the vein as following:Where O , PR and PI are the orientation, real part and image part respectively. The similarity degree of the two samples is scaled by Hamming distance.The edges of vein belongs to roof edges. The features centralize at the edges and have little relations with the central part. According to these characters, we present a vein recognition algorithm based on vector line feature. Use Canny operator to exact the vein edge points and make the energy features of 0 0 ,450 ,90 0 ,1350 orientations as the feature vector of each point. In order to decrease the dimension of the feature vector, the image is divided into blocks, make the sum of energy features of corresponding partition in each block as the whole feature vector. And in order to decrease the matching errors due to shift or revolving, we present the definition of fuzzy partition to improve the sturdiness of the algorithm. The author tested the algorithms presented in this dissertation and gave the test results. Vein recognition technology is developing rapidly and in-deep. There are yet many woeks need to be done.With the rapid development of computer industry, the capabilities of the collection devices are much higher and they become much cheaper. These support the development and application of vein recognition technology. So the transform steps of vein recognition from theory to practical merchandise will get faster and faster, the capabilities of the production will be more and more reliable, and the cost will be reduced. At the same time, because of the many advantages of vein, and the development and maturity of vein recognition technology, it will be an excellent model among the Biometric technologies.