| Significance detection extracts significant object by simulating mechanism of human visual.This technology has been applied to various fields of computer to reduce the amount of computation.So the results of significance detection can influence the accuracy of this methods.Mainstream significance object detection method get the significant figure by extraction and processing features.But both center of a priori and background a priori have the problem that the reducing of accuracy when the significant object near the border area.For the above problems,this paper presents a method on the basic of researching the relation of boundary region and central region.First of all,the image is smoothed by guided Filter.To calculate the average color distance between the center superpixel block and all the superpixel blocks,and get the significance of ordinary points by relation of boundary and center and the feature of the background to be interrupted by foreground object.Getting the first significant image by background suppression of gamma transform.at the same time,The mean value of the difference between the center significance and the superpixel block significance on the boundary serves as the threshold value to the background seed points to get the set of background seed spots.Improving algorithm of manifold ranking by the set and suppressing background to achieve the second significant image.Finally,to use pixel difference to fuse two significant images to get the final significant image.This paper is compared with seven algorithms on the data set of MSRA10 K and ECSSD.Objectively,it uses three traditional indicators,two new indicators and value of the running time to evaluate methods.Subjectively,specific experimental results were compared and analyzed.Experimental results show that background suppression,contrast,accuracy and Adaptiveness performed better under the condition of equal time complexity.The paper has 31 figures,3 tables and 54 references. |
PDF Full Text Request