Researches On Fast Independent Component Analysis Method And Its Applications Of Image Analysis

Independent Component Analysis (1CA) was a new development of signal processing. As an effective method to the separation of blind signals, ICA had attracted broad attention. Calculated higher-order statistics information, ICA could estimate the source signals, which was statistics independent and mixed by unknown factor, from the observed signals. Because ICA reflected higher-order statistics characteristic of image data, it had successful application in many fields of image processing. This paper discussed ICA's fast algorithm and its several applications in image processing:ICA algorithm and its fast algorithm (FastICA) were introduced. M-FastICA was advanced based analyzing kernel iterate course of the FastICA algorithm. M-FastICA improved convergence performance and reduced iterations. Aimed at the convergent speed of M-FastICA was dependent on initial weights, LM-FastICA was advanced by imported looseness agent, and reduced the dependence on initial weights.Analyzing the imaging mechanism of satellite multi-spectral remote sensing images, we considered the bands images of remote sensing were mixed by the spectral features of diverse surface feature randomly. The independent components separated from the remote sensing images by ICA could concentrate the surface features information, and its separability was better and could obtain better classify result than PCA.M-FastICA algorithm was applied to extract the independent component features of Yale face database images and the four normal features of CENPARMI script Arabic numerals, and the ICA features was used to recognition, and obtained better rate than PCA. On the premise of the same correct rate, M-FastICA had faster speed than FastICA.Based on the images' data and noise were mutual independent, ICA was applied to obtain the statistic information from noise-free images, and removed the independent noise from noise image using the statistic information. ICA could maintenance the original image information better than general de-noise methods.