Fuzzy Video Image Processing Based On Partial Differential Equation

In video surveillance, the captured images are usually blurred due to the weather and light changes which may affect the recognition results. Therefore, it is a key problem to find an effective method for improving the image quality in fuzzy video processing. To solve this problem, this thesis mainly studies the methods of fuzzy video image processing based on partial differential equations. The main contributions are as follows:1. The traditional image restoration methods are deeply studied, including the inverse filtering, Wiener filtering, constrained least squares method and Lucy-Richardson method, etc.. The advantages and disadvantages of each method are summarized and compared.2. Several classical variation methods for image restoration are extensively analyzed and the characteristic and numerical computing for each method are compared. Based on the analysis, the adaptive total variational(TV) model is chosen to overcome the deficiencies of the traditional total variational model. Then the model is applied to the degraded fuzzy video image restoration and satisfactory results are obtained.3. An improved adaptive TV model is proposed based on the advantages of the TV models and the isotropic diffusion model. The image edges are enhanced by using threshold and diffusion coefficient to form the fidelity. The new model can adaptively select the most appropriate restoration model based on gradient information. Experimental results show that the new model can remove noise effectively, preserve edges, and improve the peak signal to noise ratio(PSNR) compared with traditional image restoration model and the classical model of total variation. It can further solve the staircase effect in image restoration and better restoration effects can be obtained.