Research Of The Visual Attention Based On Sparse Representation And Feature Combination

With the rapid development of information technology, the amount of image andvideo data becomes larger and larger. How to finish the task of image and videoanalysis quickly and accurately has become one of the hot issues. The traditionalmethod of image and video analysis is to give the same precedence for all regions;this comprehensive processing method not only increases the complexity of theanalysis process, but also causes a lot of unnecessary waste of computing resources.Recently, many researchers have found that attention of human visual system willoften focus on several significant objects quickly in the face of complex scene. Theseobjects are processed preferentially, and this process is called visual attention. So it’svaluable to introduce the mechanism of visual attention to the image and videoanalysis task. The salient objects will be extracted as priority goal firstly for furtheranalysis. It can realize efficient distribution of computing resources as well as greatlyimprove the efficiency of the video analysis system.This paper expounds the research significance of visual attention modelingmethod and the present situation of domestic and foreign research. It also introducesthe related theory of visual attention, the modeling of attention and related conceptionof visual attention. Itti’s model and several feature integration theories are alsointroduced in brief. In the next step, we combine the sparse representation and featureintegrate theory, and introduce them into the mechanism of visual attention, then anew visual attention model based on this new method of features combination. Thispaper uses the model to analyze multi-level visual attention and simulate the shift offocus. Then, a new visual attention model for the use of dynamic scene analysis isproposed. It not only considers the static features of color, intensity and orientation,but also takes the dynamic features into account, such as spatial position, clearnessand motion information. All of these features are combined by weight, so this modelcan be used in both pictures and videos. At last, in order to evaluate the performanceof the proposed model, it is applied in several application domains, such as extractionof regions of interesting, detection of salient objects and shift of visual attention. Theexperiment results indicate the proposed model is not only sufficient but also rapidly.