Research On Theories And Algorithms For Palmprint Recognition

Biometrics is the technology that analyzes human behavior or physical characteristics for automated personal authentication. It is based on the biologic data and information technology, and has the advantages of safety, validity and ease of use. Due to its importance and promising market prospect, biometrics has been receiving more and more attention. Till now, a variety of biologic features, such as fingerprint, face, iris, voice, signature and so on, have been in-depth research and widely used.As an emerging biometrics and an important supplement to the existing biometrics, palmprint recognition has many unique advantages. Compared to the most widely used fingerprint recognition at present, the palmprint image has much larger area and more texture information than that of fingerprint. Only a low-resolution device is required for a high-precision palmprint recognition system. Compared to the most reliable iris recognition, palmprint image acquisition equipment is much cheaper. Compared to face, voice and signature, palmprint feature is more stable, reliable and hardly imitated. Therefore, palmprint recognition is valued in wider and wider range and gradually becoming to be a hot spot of biometrics.Palmprint recognition system is composed of four modules:image acquisition, preprocessing, feature extraction and matching. The image acquisition equipment tends to be relatively fixed and the recognition processes are determined primarily by preprocessing, feature extraction and matching. Preprocessing is the foundation of proper execution of feature extraction and matching. Feature extraction and matching are the keys to success in palmprint recognition. Based on the analysis of current progress of palmprint recognition technology and the evaluation of state-of-the-art recognition methods, we have investigated several problems in the the two modules to improve the precision and speed of palmprint recognition system. We propose the appropriate solutionins to these problems, which are described as follows:Preprocessing:It mainly includes alignment&normalization of geometry and grayscale normalization. The former eliminates this phenomenon of the severed finger in the binary image through thresholding and uses the continuity of finger contour to detect key points in location and segmentation. The latter takes a two-fold approach to realize adaptive illumination correction, which first removes light effect through local histogram equalization with brightness preservation and ratio method of balancing, and then denoises through hybrid spatial enhancement and hybrid denoising, respectively. After illumination correction, two kinds of images obtained achieve a certain degree of Peak Signal to Noise Ratio and a higher Edge Preserving Index, as well as conformable histograms, which benefits the subsequent palmprint line extraction.Palmprint recognition based on linear features:The improved D-S evidence theory is applied to the fusion of the ANOVA features of images from illumination correction. Linear feature extraction and ICP registration ensued. On this basis, the fusion of linear features and phase features is made on the decision level. The linear features can correct efficiently the shifting and rotation introduced in data acquisition. The phase features can keep palmprint recognition moving along smoothly. Therefore, linear features-based alignment refinement is achieved.Palmprint recognition based on coding methods and texture descriptors:The output of the2D Gabor filter is a complex number which is determined uniquely by its phase and magnitude features. Furthermore, phase features are more important than those of magnitude. Competitive code (CompCode) uses only its magnitude features. Combine the phase and magnitude features from hybrid Log Gabor and Gabor functions into a complete frequency response and joint coding them for palmprint recognition. On this basis, we build a model to describe the geometric transformation——SOG descriptors, mainly because the parameter space of palmprint images has Lie group as well as its expanded manifold structure. Optimize parameters directly on manifold, which can be solved by exponential mapping between Lie group and its Lie algebra. Correct efficiently the shifting, rotation and nonlinear deformation introduced in data acquisition leads to improve significantly the recognition rate.Palmprint classification based on a hybrid approach:For ROIs, a mixed feature extraction method of the fusion of2D Symbolic Aggregate approximation representations based on time series with competitive codes is put forward. A set of binary classifiers are obtained through ECOC.GentleBoost. Multi-class classification in the course of coarse matching not only improves the recognition rate but also speeds up the identification process.