A Study On The Hand Vein Recognition Based On The Fusion Of Pixels Annular Means And Local Binary Pattern

With the development of the times, artificial intelligence technology is used more and more in people’s life and work, which sets higher demands on the protection of information security. Under such context, biometric technology is to applied to the information security. As a kind of biological feature, hand vein has gradually become a hot research for its unique advantages, for example, intravenous information will not be copied for that it is in the back of hands, the collecting equipment costs relatively cheap and it’s acceptable.Nowadays, hand vein recognition research is still in the development stage. As for the research of the design of hand vein collecting system and the collection of the images’ region of interest, people have different methods and understandings. The most critical is the research of hand vein recognition algorithm. The traditional method is mainly using single algorithm for identification. Compared with the single algorithm, multiple algorithms have great advantages in real-time, accuracy and many other aspects. Decision-level fusion algorithm is proposed in this thesis in which two kinds of modified hand vein recognition algorithms are fused.The main work of this research is as follows:(1) The hand vein collecting system and light source system are designed. We use the transmission method to collect the hand vein so that the drawbacks of the reflection method used before can be avoided. Tests have shown that the hand vein image with clear texture can be got by using near-infrared LEDs transmission method.(2) The hand vein images collected are pre-processed, the main work is selection and position of region of interest. By comparing the existing algorithm, the author designs concrete extracting algorithm based on the specific features of hand vein images collected by using transmission method. Experiments show that this algorithm can help us get rich hand vein information.(3) Having analyzed the application of BP neural network algorithm in image matching and compared the related algorithms, annular pixel mean-matchingalgorithm is proposed on the basis of mean absolute deviation(MAD) algorithm and neural network algorithm. The experimental results show that the algorithm can identify within 15 ms, which means that it can meet the requirements of the implementation of matching.(4) The modified local binary pattern(LBP) algorithm is put forward based on the analysis of local binary pattern(LBP) and how to improve the algorithm so as to make it appropriate for hand vein. It is proved that the improved LBP algorithm recognition rate reached to 98.8%.(5) The fusion algorithm is analyzed, which is proposed by fusing the annular pixel mean-matching algorithm and the modified LBP algorithm. This algorithm combines the advantages of the annular pixel mean-matching algorithm and the modified LBP algorithm, which is proved to be of high accuracy and speed of recognition. The rate is reached to 99.2%, and the recognition time meet the requirements of real-time recognition.