Palmprint Identification Of Key Technology Research

In today's information society, there are many occasions to identify personal real identify, biometric recognition technology is the most important of solving this problem and reliable method, for personal identification,which is based on human behavior or physiological characteristics, and utilizes the approaches of image processing, pattern recognition to implement identification biometric recognition technology takes advantages of being, safe, reliable and easy to use which has attracted more and more attentions and became a new hot spot in the field of pattern recognition.Palmprint recognition is a new biometric identification technology. Compared with the fingerprint, face, iris and other biological characteristics, the palmprint posses unique biological characteristics including rich and stable texture information, low resolution and low cost acquisition and easy to locate and extract region of interest (ROI). futhermore the many research results show that palmprint based biometric recognition systems yield very high correct accuracy. Therefore, the palmprint recognition technology became a very important and active issue. This thesis addresses this issue and proposes several approaches to palmprint image preprocessing, feature extraction, and feature recognition, the main contributions of this dissertation are summarized as follows:(1)For region of interest (ROI) indicators, this papers adopts the ellipse fitting segmentation method to extract palmprint region of interest (ROI). In the method, the palmprints in the most prominent position are separated from the background using a threshold, the center of palm is determined via distance transform, and then the optimal ellipse is located, which surrounds the palm area, the palmprint is rectified using the major axis direction of the ellipse, via correlation algorithm. This method can quickly, accurately, and efficiently segment the rich information from palmprint images.(2)In order to improve the speed and accuracy of palmprint identification system and overcome the limitation that the recent statistical algorithms for palmprint recognition lack the ability of orientation selectivity and representation of the major wrinkles and principal lines in palmprint images, a novel identification algorithm for palmprint identification is proposed. In the proposed algorithm, palmprint images are normalized in the orientation, position and illumination conditions based on the integrated optical density, moments and central moments, and then one order statistics such as means and variances of each sub-band are calculated in their Contourlet domains and regarded as features. A support vector machine-based classifier is employed to implement recognition. The experimental results shows that this method is more effective in matching than Fourier transform, Wavelet moment, and Hu's invariant moment algorithm.(3)In order to achieve high identification accuracy at a low computational cost, this paper provides a shiftable and gray scale invariant description of image, a novel feature extraction framework is presented for palmprint identification. In this framework, the image is firstly decomposed by the shiftable complex directional filter bank (CDFB) transform which provides a two-dimensional (2-D) decomposition of energy with shiftable and scalable multiresolution, arbitrarily directional resolution, low redundant ratio, and low cost implementation. The subband coefficients of CDFB decomposition are operated by the uniform local binary pattern (LBP) which is gray scale invariant and contains information about the distribution of the local micro-patterns. The resultants are divided into many subblocks, and the statistical histograms of the subblocks are achieved independently. A Fisher linear discriminant (FLD) classifier is used for palmprint identification. Experiments are executed over the Hong Kong PolyU palmprint database with7752images. To verify the high performance of our proposed method, several other multiresolution and multidirectional transforms are also investigated including Gabor filter, dual-tree complex wavelet and Contourlet transforms. The experimental results demonstrate that CDFB yields the most promising performance balancing the identification accuracy, storage requirement and computational complexity.