Palmprint Recognition Based On Feature Fusion

Palmprint recognition identifies different people through extract and analyze the palmprints automatically by computer. The recognition process relates to the science of digital image processing, pattern recognition, machine vision and artificial intelligence. It has a broad application prospect in fields of work attendance checking system, access control system and safety inspection place. The research on palmprint recognition technology has a high theory sense and application value.The mainly research works are as follows:1. The preprocess methods of palmprints were researched. It includes the geometry preprocessing and the gray preprocessing. The purpose of the geometry preprocessing is to extract the palmprint region that has the most valuable classification features from initial palmprint images, and in this paper, a new convenient method was proposed, and it could cut the initial images correctly; the purpose of gray preprocessing is to change the gray value of the image, and then obtain the new image which has a higher ability to extract features, and histogram equalization, median filter, mean filter and Gauss filter methods were compared in this paper, and through analysis of the preprocessed image, the gauss high-pass filtering was chose to be the gray preprocessing method of the palmprint images.2. Four typical palmprint feature extraction methods were researched, and the Two-Dimension Discrete Fourier Transform, Two-Dimension Discrete Cosine Transform and Two-Dimension Wavelet Transform were used to extract and contrast the frequency features of the palmprint, while the Local Binary Pattern uses the airspace features. The methods mentioned above can obtain a large number of feature quantities, and the features are too meticulous, so based on them, energy features were calculated, this step can decline the dimensions of features, and at the same time, the inner features of palmprint were reserved. Experimental results show that the extraction method that has the highest rate is Discrete Wavelet Transform, and it could obtain the recognition rate of 98.6%.3. To breakthrough the limited message from single feature, this paper mainly researched the palmprint recognition method by feature level fusion. The process of feature fusion has three steps: feature choosing, weighted processing and decline dimension processing. The purpose of feature choosing is to remove some feature value which has low efficiency or disadvantage efficiency, and restrain the useful feature value; weighted processing can enhance the important features and weaken or ignore the unimportant features by give different weighted values to different features, and could obtain the purpose of improving recognition rate; decline dimension processing can decline the dimension of feature by condense features, and still have a high recognition rate. The DWT and improved LBP which obtain the first two recognition rate were chose to make the fusion experiment, and obtained the recognition rate of 99.8% at last.The feature extraction methods of palmprint were researched in this paper, and multiply feature fusion method was researched based on the above. The limit of single method could be overcome through feature fusion, and the recognition rate improved efficiently, experimental results show the effect of this method.