Research And Application Of Independent Component Analysis On Image Visual Perception

In the features of visual perception about image, color and shape features are the two basic features. It is a hot issue and also a unresolved problem that how does the human visual system perceive these features and achieve the perception of image information at last. With the information technology rapidly developed, it is very important to explore the perception mechanism of the human visual system for image color and shape features. It has very important theory meaning and application value for the promotion of recognition and retrieval to establish the recognition to solve model just like the human brain cognitive in color and shape features.Functional magnetic resonance imaging (fMRI) is efficient instrument for revealing the brain cognitive process, and the analytical methods of the fMRI data are constantly progress and improvement. Independent component analysis (ICA) in recent years developed by the blind separation is widely used in fMRI data analysis. Spatial ICA is more widely used, simply because the space dimension of the fMRI data is greater than the time dimension. But current spatial ICA method in the treatment of the fMRI data has shortcomings and the insufficiency. For instance, the nonlinear function choosed for simple calculation is not able to adapt to different source distributions; running time is not stable because of the randomization of the initial separation matrix during the process of iteration. It can lead to incorrect results for spatial ICA in the fMRI data processing. Visual perception will be researched by fMRI. It is key point to focuses on the one of analysis methods of fMRI data-spatial ICA. It is presented a new algorithm-PsICA algorithm on questions of the selection of nonlinear function and the adjustment of step length. The new PsICA algorithm can adjust adaptively nonlinear function according to fMRI data and adjust automatically step length in the iterative process, and it can ensure the effective results of the separation and can be more stable running time. It provides new ideas for the computer to process the basic information of color and shape features in the color image. Main work is as follows:1. Designning the experiment paradigm for the task of the perception of color and shape features; collectting and preproccessing the fMRI data.2. Studying in-depth the sICA mathematical model and its implementation algorithm; putting that qualifier on two shortcomings of the existing algorithm, such as:estimation of nonlinear function and automatic adjustment of step; presenting the new PsICA algorithm, and using the new algorithm to deal with the mixed data; evaluatting the performance of new algorithm about the veracity, running speed, separation effect and stability, and results showed that the PsICA can effectively separate independent components from the mixed data.3. Spatial independent component analysis based on Pearson system (PsICA) is applied to analysis the visual perception network on the image color and the shape features, and research on the function separation of the characteristics identification. Local integration is studied on the independent perception networks by using the shortest distance clustering. The results showed that visual perception system is formed with different small perception systems integration. Small perception systems have the fundamental function to recognize the characteristics of the object. Similar perception systems corresponding to the visual cortex are approximately the same areas, and activated state of the activated visual areas is depended on the specific characteristics. It provides a foundation for further study on the features bingding and certain reference value for the computer visual feature binding.