Research On Computational Model And Experiments Of Visual Attention Based On Space And Object

Vision, which is one of the most important perceptions of human beings, is an extremely complex process of perception and thinking. About 80 percent of the outside world information is obtained through the vision. At present, the human visual attention mechanism, which is characterized as the quick access to the complex outside scenes, is becoming a hot research topic. Under this topic, to set up a computational model of visual attention is critical to ensure the effectiveness of a machine vision system. The choice of visual attention units is the key issue to visual attention model.Firstly, this paper introduced the research background of the topic, the theory and the development of calculational model of visual attention. It also analysed the development of perceptual organization and the perception of the Gestalt principles of organization. For the more, the human visual information processing mechanisms was studied.Then, the image was devided into many blocks. Each block's characters, (ie the mean, the variance, the magnitude and the rate of change) were calculated. Then the method used difference operator to get the characteristic images and then ultimately merged them into a salency map. Each region's visual entropy in the saliency was calculated. The largest visual entrory of a region was regarded as the focus of visual attention. The next focus was on getting through the return of the inhibition mechanism of human visual attention. This paper proposed a visual preattention computational model based on visual entropy (VEVA).Next, this paper used Gabor receptive field of energy as a response to Gaussian differential structure function based on the direction of two semi-elliptical ring as a non-classical receptive field area, the design center receptive field inhibition and its surrounding region phase function to simulate the weight of non-classical inhibitory receptive field mechanisms. A new simulation of the visual cortex of non-classical receptive field model was presented. The model is used to detect the contour of an image and its effect was studied by qualitative and quantitative measures.Afterwards, according to the theory of topological gestalt perception, such as closure, continuity, proximity, multi-scale analysis techniques was used to contour perception. Based on the contour map, a multi-scale analysis and the smallest closed contour extraction method based on the mimiangle (MAMA) was proposed. The model can access the contour of objects in an image.Combining the VEVA and MAMA mothed, this paper can access saliency objects in the image, regard the contour of saliency objects as visual attention's units, using the inhibition of return mechanism, we can access the track of visual attention's focus. A computational model of visual attention based on space and object was proposed.Finally, with the application of parallel robot sculpture background, a binocular active vision to monitor the platform was proposed. The platform has a human binocular vision and flexible rotation function. The robot can observe the work space in multi-angle space, large-scale ations. With the application for the tool detection and stereo matching, the model can solve the problem of detecting tool rapidly and positioning tool accurately. Meanwhile, the model proposed in this paper has a good reference in the environment of other machine vision systems.