Research On Denoise And Segmentation Algorithm Of Finger Vein Image

Nowadays, the computer and network technology is developing rapidly, and how to assure the safety of both personal and corporate information and wealth through identity recognition becomes a crucial issue. Bio-recognition utilizes the innate and unique human body characteristic in the identification. Since these characters are not replicable, bio-recognition can provide more safety, security, and convenience than the traditional identity recognition method. In the current bio-recognition methods, finger vein recognition plays an important role since it is based on internal characteristics that are externally unobservable, which ensures high counterfeit-proof ability. Therefore, finger vein recognition has become a very promising research field in bio-recognition which draws great effort from researchers around the globe aiming to enhance the recognition theory and application.In the finger vein recognition system, the most important component is the image processing system. Rapidly developing together with the growing computer science technology, image processing technology has been applied to various areas, such as entertainment devices, monitor systems, medical instruments and identification systems, etc. To satisfy continuingly increased criteria in both user experience and technical parameter, certain special treatment to images are required in order to extract information more efficiently and accurately. With respect to finger vein recognition, these treatments mainly include image denoising and image segmentation. Image denoising is a key step in image processing and is often considered as a part of image pre-processing, with respect to the fact denoising should be performed before other steps at most cases. Many methods have been developed according to noise characteristics, and applied to specific areas to solve specific problems. The purpose of image segmentation is to separate target of interest from background. The output of image segmentation is the precondition of following analysis which directly influences the result of image processing. Therefore, image segmentation is the connecting step between image processing and image analysis, and is placed at a crucial position in image engineering. The research of image denoising and segmentation has great contribution to the development and application of digital image processing. This thesis is supported by the Science and Technology Development Project of Jilin Province. Major effort was devoted to the research of image denoising method focusing on complicate background noise and segmentation method for image of low quality, such as blurred edges, low contrast and partially lighted, etc. With regard to image denoising, we focused on reconstruction method based on compressed sensing; with regard to image segmentation, we focused on the local binary fitting segmentation based on active contour model. Then, we applied the research achievement to finger vein recognition system, and obtained promising results. Detailed thesis contents are:We analyzed the gradient projection sparse reconstruction method based on compressed sensing. In this method, it is not necessary to consider the noise distribution or compressed sensing of the original signal. We applied this method to the finger vein images and obtained promising denoised result. Other traditional methods, such as partial differential equation denoising, and wavelet threshold method, were compared with this method. The result indicated that this method was very effective for images with complicate background noise and blurred edges. Besides, the method showed strong performance in preventing information distortion and at the same time removing noise influence. And we analyzed the performance of gradient-projection based sparse reconstruction method in the optimizational iteration. It was noticed that convergence time could be impacted by the serrated iteration function. To solve this problem, we proposed the method that performs sparse reconstruction by using P-R conjugate gradient projection method when solving the constrained quadratic programming problem. Originally, the iteration direction of a given point is the negative gradient value in the gradient projection method. We modified the method by controlling the iteration direction according to a specific function, which constructed new searching direction when iteration points increased. The new direction was constructed in the way that it should be conjugated with the direction of previous iteration point. Therefore, this optimization process had quadratic termination property which increased the convergence speed and Peak signal-to-noise ratio.Moreover, we introduced and analyzed local binary fitting and active contour image segmentation. Based on this method, we proposed a new model, called energy functional model based on weighted total variation and local binary fitting. This model inherited the ability of LBF model, such that obtaining neighborhood information by using Gaussian kernel function and avoiding wrong segmentation caused by uneven gradient. Besides, an edge stop function was included in the weighted total variation component to enhance the sensibility to edge information and reduce wrong segmentation caused by low image contrast. Nevertheless, regularization constraint was applied in the model to prevent re-initialization of level set function and reduce the computation time.Finally, we denoised the simulated data and real finger rein images by using the PR conjugate gradient projection method. Then, finger vein area was segmented by applying the weighted total variation and local binary fitting segmentation method to the denoised images. The segmented area was processed by morphology-based skeleton method to obtain more accurate finger vein skeleton. The experiment results indicated that, the proposed system/method is effective for extracting finger vein features from vein image. The obtained vein area has good accuracy and detail information, which is of great significance for the finger vein recognition and registration.In summary, this thesis included a thorough study of theoretical study towards finger vein image recognition, presented unique innovations to recognition method, and verified the proposed method by experiments. The experiments results proved that the method was very effective for the studied topic. Based on synthesized theories, the proposed method was both comprehensive and innovative, and can be applied to other image processing area to promote individual or public information security and contribute to the information modernization of the society.