Research On Visual Selective Attention Mechanism And Its Application On Image Compression

Active vision is the hot field and the developmental direction of machine vision, in which the key problem is how to acquire visual information actively,intelligently and selectively under a given visual task.From the viewpoint of computational,it is to implement a computational model of visual selective attention mechanism to compute the saliency of image data,increase the effectiveness of computer image information processing.There are three primary aspects in the research of selective attention mechanism for image information processing:How to construct a practical framework for the selective attention mechanism? How to automatically identify the focus of attention that generally is the region of interest? How to perform the current application? This dissertation studies these aspects in details,and some valuable results are achieved, including the following research contents:Firstly,visual attention computational model based on tracking target in the video sequences is studied.In the dynamic environments,a visual attention computational model based on tracking target in the video sequences is constructed. This thesis analyses the Itti's model has the complex and slow computation in the dynamic environments,and it is not effective in real-time processing.Then the thesis uses the similarity between the adjacent frames,establishes the color histogram, selects the maximum similarity as predicable model,and gets position of the focus of attention in the next frame.Secondly,a computational model of visual selective attention deployed by depth information is studied.The model presented in the thesis aims at the bottom-up aspect of covert attention which deploy by depth information.Based on the research of neur-oscience of visual attention,intensity,color,orientation and depth information are used as the features attracted attention in this thesis.It uses Itti's model to compute color, intensity,orientation.Then based on the segment-based stereo matching which using belief propagation and a self-adapting dissimilarity measure,the depth information is computed.At last,intensity,color,orientation and depth information are deployed and create the saliency map.The saliency map is deployed to the FOA through inhibition of return and WTA mechanism.Finally,a computational model of visual selective attention deployed by depth information and image compression is studied.In order to compress images more efficiently,the thesis discusses how to combine the model and image compression.A saliency based bottom-up visual attention computational model deployed by depth information which is motivated by visual physiological and psychophysical experimental results is used,extracts the ROI.Then it is encoding based on the JPEG2000 algorithm. The ROI of the image is compressed with a low compression ratio and the background with a high one.The images compressed have perceptually high quality.The proposed approaches and algorithms are applied to the various images respectively,and the experimental results are prospective.