Study On Palmprint Recognition Algorithm Based On Wavelet Transform And Implement On DSP

Palmprint recognition is a new technology in the field of biometric identification. It is use of an effective area on the palm of your hand and extracts the characteristic information of palmprint. according to the information to identify users with these features. Currently, palmprint recognition in the entire biometric market share is also low. Palmprint identification products for civilian are seldom. Therefore, the development of small size, low cost, shorter development cycles, easy maintenance, to meet the medium and low-level secure palmprint identification system in order to promote civilian areas is a subject of great research value.There are many palmprint recognition algorithm, some of which are easy to implement on DSP, such as extraction method based on the main line, image-based transform method. In view of the main line extraction method is time-consuming when it is matching and the recognition rate is not high. In this paper, we choose the method based on image transformations. According to the multi-resolution characteristics of wavelet transform, this paper choose to use 5/3 wavelet transform to extract the palmprint features.In the preprocessing stage, in order to extract the smooth palm edge, using the Gaussian low pass filter eliminate noise to the input palmprint image , and then processing palmprint image by binaryzation, at the same time, using improved 8 direction template prewitt operator extract the edge of the palmprint image. At last, we get a good edge effect. On this basis, the method of digital morphology is used to extract palmprint image corners region, and with the thinning algorithm to get a single pixel corner , Thus the effective area is intercept from the palm.In the recognition phase, Using 5/3 wavelet, the effective area of the palm were decomposed to third level, given the definition of wavelet energy, taking into account the texture characteristics of palm, the energy of principal lines and wrinkles, which are nonoscillating patterns, are concentrated at the large wavelet decomposition and the most energy of ridges, which are oscillating pattern. not a shock of palm lines are concentrated in large-scale energy details of the wavelet decomposition of the image, are focused at the small scales. Therefore, the detail image were divided into equal blocks after third level transformation, the detail were divided into 12 blocks in each level, and finally, construct the wavelet energy feature. To take advantage of the constructed wavelet energy feature calculate the similarity of two palmprint images. To test the recognition accuracy, in this paper, the extracted palmprint features were registered, and then using a registered palmprint to match a under test one. Therefore , we can calculate the recognition rate of the palmprint recognition algorithm.The hardware platform is the DM6446 in this paper, in the reading of palmprint image, memory allocation, recycling pipeline control, and at the aspects of optimizing code writing. After code optimizing, The recognition program running time reduced from 7.6 seconds before optimization to 1.6 seconds. The optimization effect is obvious ,which can meet real-time requirements.