The Research Of Palmprint Identification Algorithm

As a novel biometric identification technology, palmprint recognition is a hot research area in recent years, it has the advantage of convenient image acquisition compares to hand vain recognition, rich feature information compares to hand shape recognition and high acceptance level compares to iris recognition. But at present, palmprint recognition is still in the underway phase, there are a lot of questions which are needed to be addressed at each stage, and this thesis carries out the research in the above background.Our research consists of three parts:the extraction of region-of-interest (ROI) in palmprint image, the enhancement of palmprint image and the feature extraction method of palm print image.Region of interest (ROI) extracting in palmprint image based on the curve of gray-level integral projection. This method is proposed specific to a particular database named PolyU Palmprint Database, after a global thresholding, we applied the curve of gray-level integral projection to palmprint image to detect the key points which used to calibrate the direction of palmprint image and build reference frame, finally we cut a region size of 128*128 as standard palmprint image.Enhancement of palmprint image based on sobel gradient operators and unsharp mask in the fuzzy domain. In order to eliminate the effects of the tiny gray-level difference made by image acquisition, we did gray-level normalization at first, and after that enhanced the palmprint image using method proposed in this thesis. The first step of enhancement is sobel gradient operation, after this step we got a principal lines image, then adopt average filter to smooth principal lines image; furthermore, with the purpose of enhancing contrast between principal lines and skin, we used a half-opened membership function to map the image to a fuzzy domain which range is between 0 and 1. Then we finished enhancement using unsharp mask algorithm. The enhancement improves the contrast between principal lines and skin region in the palmprint image; eliminates tiny shadow caused by hand incorrect placing or not completely stretching as well. Finally, to make principal lines more obvious, we did gray level stretch on the image using Cosine gray stretch function.Feature extraction method of enhanced palmprint image based on two-dimensional real discrete Gabor transform (2DRDGT). At the very beginning, palmprint image are divided into 8*8 blocks that are no overlaps and applied 2-DRDGT in each block, got Gabor coefficient size of 8*8*16*16. Then, we regarded 64 blocks’energy as palm print feature; the experimental result based on Euclidean distance classifier and a 600 images database which includes 30 different classes obtains the recognition rate of 97.56%, then we built a PCA transform matrix which is used to act on coefficients of each block in order to reduce their dimension into 256, using this vector as palmprint feature and based on the same database gets the recognition rate of 99.05%, recognition time for each image is 0.41 second approximately which is in an acceptable range.