Hand Vein Recognition Based On Feature Coding

After the September11terrorist attacks in the United States since2001, information technology has been developed explosively with the information security requirements continuing to improve. Biometric technology is considered to be a reliable, stable identity recognition technology. In recent years, it has become a hot research field in information security and been paid increasing attention. The technological achievements have been successfully used in access control, security, financial, and anti-terrorism. Hand dorsal vein recognition technology, as an emerging biometric authentication technology, began in2000years ago and became the new hot spot research in the field. Compared to the traditional face and fingerprint authentication technology, hand dorsal vein has many advantages feature such as uniqueness, stability, hard to copy and forgery. However, just like face, fingerprint and iris recognition, the promotion of a wide range of hand dorsal vein recognition technology needs continuous basic research and exploration.This paper focused on hand vein region of interest extraction, feature extraction and feature encoding. First, we used image acquisition equipment to collect hand vein image database that contains a number of knuckles. Then we used location information of knuckles for hand vein image registration and region of interest extraction. Gabor, SIFT and LBP algorithms were adopted to extract the dorsal hand vein characteristics and their performance were also compared in this paper. Besides, we selected BP artificial network as an encoder for feature coding. And correlation classifier was proposed to be the final decision classification.The major innovations of our work can be summarized as follows:1. Knuckles location information was proposed to be used in the hand vein image registration and dorsal region of interest extraction. First, we located the position coordinates of the first and fourth knuckles to determine the angle of rotation. We used this angle to complement hand vein image registration and dorsal region of interest extraction. Average correlation of hand vein image was proposed to evaluate the performance of two different methods of region of interest extraction-centroid strategy and geometric strategy.2. Feature coding strategy was proposed in our work. The basic idea is to further increase distance between different classes and reduce distance within same classes. We selected a three layers BP artificial neural network as the encoder in this article. Orthogonal gold code was generated as the training target code words. Both single encoder and combination encoder strategies were proposed for feature coding. Those two strategies were also used to be compared. Combination encoder was raised in order to further reduce the error code rate. Considering characteristic of the output code sequence, correlation classifier was proposed to achieve a final decision classification.