The Video Saliency Extraction Technology Based On Temporal And Spatial Contrast

In recent years,with the rapid development of technology,both image and video data are exploding at an explosive rate.In order to quickly process a large amount of image and video data,and quickly extract useful information,Researchers are increasingly focusing on the study of significant extraction techniques.The purpose of saliency detection is to detect an image or a region of a video in which a significant target is located.According to the human visual mechanism,the saliency area is the area most concerned by the human eye.Generally,the more prominent the area is,the more value it will contain.saliency detection is widely used in the fields of image segmentation,target recognition,image compression,and image retrieval.For image and video,an image saliency extraction method based on frequency domain and objectness is proposed and a video saliency extraction method based on significant continuity and graph theory is proposed.The first is an image saliency extraction method based on frequency domain and objectness.The method first performs a curvelet transform on the input image,then reconstructs the frequency domain feature map by curvelet transform reconstruction,and then extracts the objectness feature map of the input image,followed by Superpixel segmentation of the input image,according to the customized consistency strategy,first compare the feature superpixel with the objectness superpixel,and then compare the feature graph with the objectness graph to obtain the best feature superpixel and the best feature map,and finally the final image saliency map is generated by the customized fusion strategy.The image saliency extraction method proposed in this paper has shown good results on both MSRA1000 and CSSD databases,and the selected target area is more suitable for human visual system.The second is a video saliency extraction method based on significant continuity and graph theory.The method is divided into two modules,the first module: firstly,the motion vector field is calculated by the optical flow estimation method to establish the association of the superpixels between the frames,and then the saliency is propagated to the current frame through the forward and backward directions;the second module: first The time-space gradient map is extracted,the motion histogram is extracted by the optical flow field,the static image saliency is extracted by the image saliency,and the background super-pixel fusion of the front and back frames and the current frame is used as the background prior of the current frame,and then the saliency is calculated based on the graph theory.Finally,the results of the two modules are linearly weighted and combined to generate the final video saliency.The video saliency extraction method proposed in this chapter has carried out a large number of experiments on SegtrackV1 and SegtrackV2 databases,which verifies the applicability and effectiveness of the algorithm in this chapter.