Research On Image Processing Base On Independent Component Analysis

Independent Component Analysis is a new theory developed with the field of signalprocessing in the last few years. The already wide range of applications such as imageprocessing, communications systems and signals, and voice recognition. It uses the a prioricondition of statistical independence of source signals easily satisfied, to solve the problem ofblind source separation. Independent component analysis algorithm is better calculations ofhigher-order statistics from the observed signal to estimate the implied independent sourcesignals. Algorithm to reflect the higher-order statistical characteristics of image data, so awide range of applications in image processing.This paper mainly focus on the independent component analysis study in digital imageprocessing applications. The traditional image denoising method is already quite mature, butin the noise satisfy certain conditions, it will be sensitive to pulse function, lead to separationof the denoising of images is not ideal, or it may make the algorithm does not converge.Using independent component analysis method can get better results than the traditionalGaussian method. Therefore, do the following work:(1) describes the independent component analysis of the basic concepts, principles, andare closely related to probability theory and mathematical statistics in the basics, and alsopresents a selection of independent component analysis of the independence of the measureand the objective function.(2) describes the traditional image denoising method, through the platform of MATLABsimulation and PSNR traditional image denoising method to compare the respectiveadvantages and disadvantages, which we are going to study the independent componentanalysis Denoising Algorithm a reference to the premise.(3) The image denoising method based on independent component analysis, andhighlights the Infomax algorithm and the FastICA algorithm.(4) On the basis of the original independent component analysis algorithm, theintroduction of the Kernel-ICA algorithm, the Kernel-ICA algorithm to analyze the steps.Then using the Newton iterative method to optimize the objective function, minimize theobjective function, which has made appropriate improvements to the KICA algorithm. Finally,simulation experiments to illustrate the advantages and feasibility of the KICA algorithm andimproved KICA algorithm.