Multi-modal Biometric Algorithm Research Based On Hand Image

With the rapid development of modern science and technology, there is a growing emphasis on information security. As a higher safety factor practicality technology, biometrics has been more favored by people. Simultaneously, people also put forward higher requirements on biometric technology. In all biological features, biometrics based on hand information has the highest level of acceptance among users. Palmprint and fingerprint hand information have the unique advantages of information-rich and stable feature information. In addition, hand shape has a high recognition efficiency. So based on the integration of these three characteristics we can get a biometric technology with high efficiency and high accuracy.Study on this subject has a great potential research value and broad applicability. Since the fingerprint features have been studied quite mature, the main in-depth researches are focus on hand shape and palmprint recognition technology in this passage.In the hand shape image feature recognition, the accuracy of the hand feature points location affects the correctness of the hand matching directly. On the basis of the existing hand features location, this passage proposes a feature points location algorithm, which is based on partial blocks scanning. The algorithm starts from the middle finger peak which is the easiest to locate. Then we determine the coordinates of the rest fingers peak points and valley points, step by step according to a fixed order recursive formula block based on the coordinate of the middle finger peak. Then, we extract the hand feature vectors based on the determined feature points and use matching algorithm based on the hand feature vectors to match. Experimental results show that the success rate of the algorithm's feature points is up to 94.8%. This indicates the algorithm can accurately locate the hand feature points and the algorithm is feasible. However, hand opening degree has a great impact on disc algorithm,resulting in the disk radius size and pixel threshold difficult to determine. So, the algorithm is not feasible in practical applications. In response to this limitation of disc algorithm, the author proposes disc extremum algorithm through analysis of the disc algorithm principle and hand-shaped profile characteristics. The algorithm uses the disk neighborhood extreme value method to determine the peak point and the valley point of the finger, avoiding the use of pixel threshold, while giving the disk radius a greater applicable space. The algorithm solves the limitation of disk algorithm. Experiments show that the algorithm features localization success rate is up to 98.4%. The algorithm can accurately locate feature points, so the algorithm is feasible.Because of the anchor points is not easy to identify and the deviation of extracting similar image ROI is a big problem in palmprint image feature recognition, this paper presents a new ROI segmentation algorithm. It represents the application of mathematical algorithms to find the exact points of the palm of the two valleys, with the two valleys and the palm side fitting line as a reference to establish coordinates in a fixed angle to find palm print information-rich areas and determine the image region of interest. Experiments show that using this algorithm can be more accurately to find the ROI area, and the segmentation of the same image shift is extremely small, the ROI extraction rate is up to 98.2% and the palm correct identification rate is increased by about 25% . So the algorithm is feasible. For palmprint identification we performed recognition simulation based on pixel area, wavelet transform and two-dimensional Gabor features. The results found that the effect of palmprint recognition based on two-dimensional Gabor feature is the best, not only high efficiency but also high accuracy.In terms of multi-modal integration, through research and analysis the way of multi-modal integration, we present the idea of hand features multi-modal integration in the matching layer and decision layer for hand information feature.