Probability Graph Model Based Visual Attention Mechanism And Its Application

Psychology study found that, when facing a complex scene, the human attentionwill quickly focus on a few significant visual objects, and the human visual systemwill priority process these visual objects, this process is called visual selectionattention mechanism. If this mechanism is introduced into the field of image analysisor computer vision, the computer image analysis will be significantly improved.In recent years a large number of researchers dedicated to the modeling of visualattention mechanism, and made some progress. The modeling of visual attentionmechanism has two problems: the calculation of fixation and the simulation of humansaccadic scanpaths. So far, researchers get a lot of research results on the calculationof fixation, but the simulation of human saccadic scanpaths is just the opposite.Unfortunately, there is still not very good solution. In this paper, we have a try on thetwo aspects. The thesis focuses on the following aspects:(1) We proposed a visual attention model based on Probability graph model. Inthis model, the saliency is measured by the expected of steps from the most salientpatch to every patches of a random walker. We test our model on several imagedatasets and video clips by using different measurement criteria. The fixationspredicted by the proposed model are more consistent with human fixations than someother state-of-the-art attention models.(2) A model simulating of human saccadic scanpaths based on visual fixation isproposed. Firstly, we calculate the saliency map of the input image. Then, we use theWinner-take-all to get the current fixation. The transfer process of fixation is drivenby visual memory and the forgetting property. The test result shows that the proposedmodel achieves the better prediction accuracy on simulating the human saccadicscanpaths, and it is better than some existing method.(3) We put the saliency map in content-based image retargeting. The result issignificantly better than the original results. The method using saliency map is betterable to keep the image in the relative importance of the area. In addition, visualattention mechanism is applied to the visual training with amblyopia children. Wedevelop some visual training games. These games significantly improve the visualtraining effect.