The Research Of ICA And ICA’s Multilayer Expansion Algorithm

In recent years, Independent Component Analysis algorithm (ICA) has beenwidely used in image research and application fields. It has many very goodapplications such as in the speech signal separation, image signal denoising, faceimage recognition, financial data analysis, medical signal processing, etc, these showthat the ICA algorithm’s application value. This paper is to explore the ICA algorithmand its application, as well as multiple ICA algorithm and its application research.This paper systematically expounded the Independent Component Analysisalgorithm, especially the Fast Independent Component Analysis algorithm (Fast ICA).In this paper, the application researches of the Independent Component Analysisalgorithm are all based on the Fast Independent Component Analysis algorithm. TheFast ICA algorithm has qualitative improvement in the speed aspect comparing withother Independent Component Analysis algorithm, this paper took full advantage ofthe Fast Independent Component Analysis algorithm in related application researches,such as image separation, image recognition, image reconstruction, imagecompression, which have achieved very good results.This paper extended the single layer ICA algorithm to obtain the multilayer ICAalgorithm, and applied it in the natural image, extracted image features in a higherorder，and applied them to assist underwater robot navigation and obstacle avoidanceaccording to the active image’s material identification. The core idea of Multiple ICAalgorithm is based on nonlinear transformation of the output of the single layer ICAalgorithm, so that the first layer’s output after the transformation can meet anothersingle layer ICA algorithm’s input, so as to establish multiple ICA algorithm, and toextract nonlinear high order statistical characteristics that the only layer ICAalgorithm has no way to get. As the multilayer ICA algorithm was applied to naturalgray image, it found that the image features got through the second ICA algorithm showed better texture characteristics and large area of the outline, and for the purebackground such as the sky and water can show more clear outline, and the for thecomplicated physical such as stones showed less clear region boundary. The imagefeatures got by the third layer ICA algorithm were similar to the second layer, butbecause of the higher layer, the less information to get, so features got by the thirdlayer ICA showed weaker effects than the second layer.Finally, this paper, through a large number of MATLAB tests, put the theoreticalalgorithm into good simulation application, from the experiment results, each ICAalgorithm application research can all obtain good results, so the image applicationresearch of the single layer ICA algorithm and multilayer extended ICA algorithm areall effective.