Research On Applications Of Visual Saliency In Video Object Recognition

As one of the core tasks of intelligent video surveillance, object recognition faces many difficulties. Inspired by the selective attention mechanism of human visual system, in the process of feature extraction, focusing attentions on regions that attract visual attention can save computing resource effectively as well as improving the real-time performance of the system. Based on the principle of visual saliency, this paper studies the problem of object recognition in video surveillance:First, on the basis of the selective attention mechanism in human visual system, this paper discusses the difficulties existing in object identification problem and its research significance; after discussing the biological principle of visual selective attention, two calculation models of the visual saliency are well learned.Then, by introducing the concept of superpixel and comparing several superpixel segmentation methods, this paper proposes an improved method of SLIC(Simple Linear Iteration Clustering)superpixel segmentation method. In this paper, the Sigma filter is used to filter the errors in the iterative process, which can effectively eliminate the error propagation. Next, in the view of extracting visual salient region, this paper proposes a new algorithm based on inter-superpixel comparison. A quantification treatment in HSV color space is performed on the target image. After that, by calculating the inter-superpixel difference between the center superpixel and its surroundings, the color visual saliency map of color is created; then we calculated the gray level co-occurrence matrix for each superpixel to extract its texture features, and through inter-superpixel comparison, the visual saliency maps of texture is extracted; At last, the two saliency maps are fused to complete object visual salient region extraction. Experiments show that this method can quickly and effectively extract the salient region of the target image. Finally, the visual salient regions are described, while the local ORB features are also extracted, and then the target recognition is performed by using the method of feature matching.The experiments show that the method used in this paper can improve the accuracy of object recognition, which is of great significance to intelligent video surveillance, military reconnaissance and industrial production. However, there is still a long way for computer vision to achieve the accuracy and efficiency of the human visual system.