Study Of Dorsal Hand Vein Recognition Based On Sample Quantity

Recent years, both theoretical research and commercial application about dorsal hand vein recognition technology are developing rapidly. However, there are also some shortcomings, most of the research and products are based on small sample database, there is little research pay attention to dorsal hand vein recognition under big sample conditions. Because of China’s large population, frequent exchange of personnel, current research results may not meet the needs of the crowds’ verification, therefore, the study of dorsal hand vein recognition under big sanple conditions is necessary and important for promoting the development of dorsal hand vein recognition technology.Contents of this paper are as follows:(1) The foundation of the underlying database about dorsal hand vein. We collect 102 dorsal hand vein images. By image preprocessing, including angular correction, ROI extraction, image denoising, image normalization, etc., we filter and remove the overexposure, invalid or blurry images.(2) The synthesis of the large sample dorsal hand vein images. First, the sample is divided into the characteristics sample set and mapping sanple set, then we analyze the sample by use of PC A, extract feature and structured features space. Once more, the mapping sample set is mapped to the features space to obtain the projection space. Finally, we use the principle of PCA reconstruction to realize the dorsal hand vein image synthesis.(3) The study of dorsal hand vein image separability. The self-information entropy reflects the richness of the information contained in the image. The mutual information entropy reflects the similarity of the image structure. This paper studies the separability of large sample dorsal hand vein images by the self-information entropy, the mutual information entropy and the average distance of image features between the classes and within classes. We simultaneously use traditional recognition algorithm like LBP, PCA, SIFT to sum up image recognition rate under different sample quantity and quantitative analyze whether synthetic dorsal hand vein image can be used for classification.(4) The result of experiments. With the gradual increase of the sample quantity, the dorsal hand vein image separability and the recognition rate are gradually decreased. For the dorsal hand vein recognition under large sample condition, we expect when the sample quantity increases, the recognition rate decreases slowly and can remain relatively stable. For this purpose, this paper proposes one dorsal hand vein recognition algorithm based on the deep learning and multi-scale code combinations. Compared with the traditional recognition algorithm, the results show this algorithm is effective enough for the large sample dorsal hand vein image recognition.