Study On Visual Attention Model And Its Application In Target Detection

The visual attention model, which is based on studies and hypotheses on anatomy, neural psychology and epistemology, uses math model to simulate human visual system. It has become a hot topic in the digital image processing field. The visual attention model detects regions of interest by detecting prominent points in an image, for it takes regions of interset, which contain potential target as the muster of points that have prominent features. In image processing tasks such as image retrieval, target detection and surveillance, it would be more convenient to focus on regions of interest detected by visual attention model than focus on the whole image.This thesis studies on bottom-up visual attention model. Strategies of the detection of primary visual features and combination of conspicuity maps are investigated and improved. This thesis first introduces visual attention mechanism and progresses that have been made in the field. Physiology and psychology studies related to visual attention mechanism and visual attention model proposed by Laurent Ltti are also introduced. Then an improved model is introduced. Since there are some cells in the visual system which react to the texture information, the texture coherence feature is added into the the detection of primary visual features as a kind of the primary visual feature in the new model, which makes use of intensity, colour, orientation and coherence information to form conspicuity maps. The conspicuity maps are combined as saliency map by a global enhancement method to guide attention and detect regions of interest. Experiment results have demonstrated that the improved model can detect regions of interest with strong coherence feature which fail to be extracted by Itti model, and regions of interest are detected in a more reasonable order.This thesis also proposes an algorithm to detect moving target based on visual attention. The detection of moving target is an important field in image processing and computer visual. This thesis first adopts the visual attention model to detect the regions containing potential targets, then use invariant moments to find the best match. Experiment results on autocar video frames have shown that on the premise that the moving target is not covered and does not have non-linear distortion, the algorithm proposed by this thesis can detect the moving target effectively.