Research On Image Signal Processing Method Based On Volterra Filter

Digital image processing generally uses a method of computer image analysis and processing, to meet the visual needs of human eyes or other devices. Currently Digital image processing has become more important in the fields of computer science, medicine, information science etc., and has great value in the theoretical study and practical applications. Image denoising is a basic technology in kinds of image processing technologies, which has great significance because image denoising is the preprocessing step of other further image processing. In the field of image denoising, due to the linear characteristics of linear adaptive filter, the nonlinear signal higher-order redundancy and nonlinear function approximation are restricted, but the Volterra filter considers linear characteristics and nonlinear characteristics thoroughly, when the condition of limited energy systems is satisfied, Volterra functional series can reach the accurate approximation of the vast majority of non-linear systems arbitrarily. Thus, research on Volterra filter theory has very important theory significance and application value. According to several kinds of common noise in image processing, this papar proposed several novel image signal processing methods based on Volterra filtering combined with LMS and RLS adaptive algorithm, and achieved some valuable research results.Firstly, this dissertation introduced the basic theory of image filtering and Volterra filter. The section of image filtering theory expatiated the common types of image noise, the classic image filtering method, and the performance evaluating indicators mainly used in image filtering method. The section of Volterra filter theory expatiated Volterra series, common nonlinear filter, Volterra series characterization of nonlinear systems and specific Volterra filter model.Secondly, this dissertation studied the Volterra image filtering methods based on the LMS algorithm. In order to improve the Volterra filtering performance and applicable range, this dissertation puts forward three improvement methods based on VLMS. The first one, for the short comings of VLMS filtering algorithm which converges slowly and is difficult to determine the step factor, furthermore, its convergence rate is influenced by the eigenvalue of input signal correlation matrix. This dissertation proposed the image Volterra filtering algorithm based on normalized LMS (NVLMS), through normalizing the step length factor by2-norm of the input vector to replace the original fixed step length, increased the rate of convergence of the algorithm and the image filtering effect is improved. The second one is considering the a stable distribution impulse noise environments, this dissertation discussed the degradation of second-order moments based on VLMS and NVLMS algorithm, and proposed the image Volterra filtering algorithm on LMP (VLMP) based on minimum deviation criterion. The VLMP algorithm extends the application range of Volterra filtering algorithm, but the algorithm is only suitable for (1≥α≤2) of the a stable distribution impulse noise environments. The last one, this dissertation proposed the improved VLMS algorithm (M-VLMS) image filtering algorithm, which uses nonlinear transformation to suppress the impulse noise effectively and improve the filtering effect. Theoretical analysis and simulation results show that the proposed three improvement methods, NVLMS. VLMP. M-VLMS, have good filtering effects and edge protection capabilities compared to the mean filter and median filter algorithm under Gaussian noise environment. While under the a stable distribution impulse noise environments, the performance of NVLMS algorithm degrades, and the performance of VLMP algorithm has improved compared to NVLMS algorithm, but the effect is not ideal and the algorithm is not suitable for (∝<1). However, M-VLMS algorithm has better filtering effect in the whole range (0<a<2), and the algorithm also has the better image denoising performance under high density random value impulse noise environment.Finally, according to the characteristics of slow convergence speed and the filtering performance of LMS is to be enhanced, several new improved image Volterra filtering algorithms are proposed, including the image Volterra filtering algorithm on RLS (VRLS). the image Volterra filtering algorithm on RLP (VRLP) and the improved VRLS image filtering algorithm (M-VRLS). Theoretical analysis and simulation results show that under Gaussian noise environment, compared to mean filtering, median filtering and M-VLMS algorithm presented in chapter four, the filtering effect of VRLS algorithm has been significantly improved, and is suitable for different densities random value impulse noise environments. However, VRLS algorithm degrades and even is completely failing under a stable distribution impulse noise environment due to the second-order statistics filtering method Therefore, the performance of VRLP image filtering algorithm is improved under a stable distribution impulse noise environment, and which is suitable for Gaussian noise too. But VRLP method is also influenced by value range of p parameter, only applies to the range of1≤α≤2. M-VRLS image filtering algorithm based on the fractional lower order covariance deals with the input image signal using a nonlinear preprocessing method, which can achieve the image denoising and image detail reserved effectively while suppressing the impulse noise. M-VRLS algorithm has remarkable improvement of filtering effect in both Gaussian and the whole range(0≤α≤2)of the a stable distribution impulse noise. Its performance is better than the mean filtering, median filtering algorithm, M-VLMS algorithm and VRLS algorithm, while M-VRLS image filtering algorithm also has optimum denoising effect under different densities random value impulse noise environments.