Study On Video Saliency Detection AlgorithmBased On Motion And Spatial Direction

Visual saliency is based on the characteristics of human visual system. By establishing the visual saliency model, we can quickly find the focus area in the image for various applications, ROI detection, target recognition, tracking and so on. In some video scenes, the moving salient object detection may be affected by the moving complex texture region in background. Video saliency detection algorithm is studied in this paper in order to solve the problem of the background texture region movement.A video saliency detection algorithm based on motion and spatial direction is developed in this paper. The characteristics difference of the complex texture area in background and the salient motion area in the video image is studied and analyzed. For three cases of background motion, a 3d space W is built, which has three axis of horizontal gradient, vertical gradient and the amplitude of motion vector.3D structure tensor matrix is used for combining the information of motion and spatial direction in video image. Then the eigenvalues of the matrix are calculated to get the intensive degree of the corresponding points of each pixel and the surrounding pixels in space W. By post-processing the intensive degree, the video saliency map is obtained.The performance of this algorithm is optimized. First, the building 3D structure tensor module is improved, which can effectively reduce the number of multiplications and additions. Experimental results show that this improved method can reduce 83.47% of the running time of the algorithm in this paper. Then parallel algorithm is implemented. Data parallel structure is used to complete the task of building 3D structure tensor matrix, which has the longest running time. Experimental results show that the 4-core parallel speedup of this method is up to 3.43.The accuracy of the algorithm in this paper is tested and analyzed.5 video sequences are adopted in this paper, including all kinds of motion cases in background. Then the test results of the algorithm in this paper are compared with those of the three classical algorithm. Experimental results show that the accuracy of the algorithm in this paper is better than that of the classical algorithm. The algorithm in this paper can eliminate the influence of the moving complex texture region in background, and can detect accurately the video salient region.At last, the topic of this paper is summarized and the future research direction for further improving the accuracy and the performance of the algorithm is prospected.