Research And Realization Of Knuckle-print Identification Method

Biometrics uses the unique physical or behavioral characteristics of a human for individual identification through computer, biosensor and principles of biostatistics.The study of new biological characteristics can make up the existing defects of the existing biometrics technology. Knuckle-print is a new characteristic which has the same stability and distinguishability as the other hand characteristics. Therefore, we use knuckle-print for identify recognition and research the key technology of identification process.The purpose of hand image preprocessing is to obtain the Region of Interest (ROI) based on the reference coordinate system which is established by hand corner points. Preprocessing is necessary because it can improve the system performance and robust of algorithm. Feature extraction is the core of pattern recognition. It is important to select certain feature from high-dimensional feature space for classification and match of knuckle-print. This paper has researched the preprocessing methods and feature extraction algorithm according to the knuckle-print features. Research of the dissertation is concentrated on the following aspects:(1) A novel binarization method is proposed in this paper. We use the HSV color model to reduce the uneven illumination of gray level threshold binarization method. It can segment the hand from background effectively.(2) The Harris corner detection algorithm is improved in order to detect key points of finger-tips and finger-roots. The ROI is obtained after reference coordinate system is established.(3) Use 2D Gabor filters filer the ROI of knuckle-print and the binary real part and imaginary part as knuckle-print feature. Take the XOR operator to finish the knuckle-print matching.The proposed algorithms can effectively implement knuckle-print identification and verification. It is shown by experiments which proceed in our own building image data and the data provided by Information Institute of Beijing Jiaotong University.