Research on Visual Attention Model with Joint Spatial and Feature Domain Information
|School||Beijing University of Technology|
|Course||Computer Science and Technology|
|Keywords||Visual attention Dissimilarity Spatial distance Central preferences Image scaling|
Note that the selection mechanism is an important characteristic of human visual perception , if the mechanism of the human visual system is introduced to computer vision system is bound to enhance the performance and efficiency of existing computer image analysis . Way to build their applications in order to explore the computational model of the human visual attention mechanisms , specific research work as follows : First, the paper reviews the bottom-up model of visual attention , since the top-down visual attention model and viewpoint transfer mechanism Research also describes the visual attention model in image segmentation, image classification and target detection . In-depth study of human attention mechanism , this paper analyzes the analog visual attention of three key factors : the spatial distance between the image dissimilarity between the local area , the image local area , local image regions to the center of the image distance central preference characteristics . Secondly, a joint airspace and the feature domain information visual attention model , the model on the basis of the three key factors , core step is the spatial distance between the image local area and central preference characteristics of the local area dissimilarity between modulation , and its essence is the image of each region of the human eye is defined as the degree of concern based on the spatial distance weighted global features response rarity . Multiple image libraries and video clips on different testing standards , test results show that compared with some well-known international attention model , more consistent with the viewpoint of the model predictions with real human viewpoint . Finally, the model is applied to content-based image scaling. The subject of the image is generally reflected in the character, nature, landscape , urban landscape semantic information , the lack of detection of this type of information the model , this paper on the basis of the model add a face , pedestrian and vehicle detector . The test results show that compared with the conventional image scaling method of the model - based approach , the foreground object in the image distortions can guarantee smaller .