Video Saliency Detection using Weight Based Spatio-Temporal Features

The issue of enumerating the dependability of computational saliency in a video, which can be utilized to propel the processing algorithms used for saliency-based video calculation and unwavering quality of the procedure are tended to in this module. Weight based spatio-temporal approach is used to enumerate such reliability in twofold: 1) investigating the spatial connections in both the saliency graph and the eye-fixation map.2) taking in the weight-average based spatiotemporal relationships, which characterize a dependable saliency outline. First examination on spatio-temporal eye fixation information from the general public CRCNS dataset and examination of a typical element in human visual consideration is done, which ascertains the relationship in saliency inside a pixel and its immediate neighbors. This algorithm estimates a pixelwise uncertainty map that replicates the assurance in the associated computational saliency map by relating it to the pixel's saliency of its direct neighbor. The difference of a pixel from its nearby neighborhood is estimated in the saliency graph to gauge such vulnerabilities. Then we add a weight value to each pixel, thus producing more reliable output. The objective of this module is to produce more reliable saliency map. The simulation tools used for this research modules is MATLAB.