Multispectral Palmprint Recognition Method Research

Along with the rapid development of information technology, people are concerned with information safety and public safety. As a result, biometrics was invented accordingly. Biometric is a kind of identity verification technology using the physiological and behavioural characteristics of human as the mean for identification. Finger print, palmprint, retina, iris and face are widely used features. Among them, palmprint has received great attention due to its easiness of collection, low cost of sample collecting equipment, high user acceptance, easy extraction of important features, etc.Multi-spectral palmprint recognition technology utilizes the absorption and refraction characteristics of light at different wavelengths at the palm surface and the tissue underneath. More than one palmprint images can be obtained from a single palm, those images are then combined to significantly increase the recognition rate. Two important techniques are used in this technology:(1) Image feature detection (2) Image fusion. In this paper, several crucial techniques are used, such as image pre-processing, feature detection, feature storing, feature matching and score level fusion. With a focus on the features of the palmprint and palm veins from multi-spectral palmprint, crucial techniques and core algorithms of the recognition technology are investigated. Research includes(1) Technique based on the geometric features of a palm, using the troughs between fingers as the reference to establish a coordinate system, and thereby separating the region of interest (ROI) from the palmprint image.(2) Down-Sampling method with dual-cubic interpolation algorithm.(3) Using the multi-resolution and multi-directional characteristics of Non-subsampled Contourlet Transform (NSCT) to detect features from the ROI of the palmprint image, and therefore, the direction map for feature encoding. At the same time, analysing the effect of the size of ROI image and the number of decomposed direction on the performance of NSCT feature detection.(4) Encoding direction feature by fuzzy hashing function to obtain the binary hash table of the phase-position and direction features for feature matching. (5) Finally overall recognition rate is deduced by merging the matching scores from all the RO1palm images. Large amount of experimental results demonstrated that multispectral palmprint recognition technology can achieve high recognition rate. When the Equal Error Rate (ERR) is0.16%, a high recognition rate of99.83%is obtained.