Research On Saliency Detection Based On Background And Foreground Seeds

In recent years,many experts are based on exploring saliency detection.The main purpose of the image saliency detection algorithm is to identify salient objects,that is,the longer the human eye stays in the line of sight,the more interesting the area.In addition,we must spare no effort to eliminate other complicated information.The final saliency map is obtained through the saliency detection algorithm,and a grayscale image is displayed in front of people.The higher the saliency value of the pixels in the image,the greater the probability of being considered as a saliency area.Although the algorithm of saliency detection from simple to complex has achieved good experimental results,it still has great room for improvement.To this end,this paper proposes a new saliency detection algorithm,the specific work is as follows:(1)Provide new ways to acquire background and foreground seed sets.This method uses edge strategy to eliminate foreground pixels mixed with boundary seeds,so as to obtain background seed sets and background saliency map.The adaptive threshold segmentation technique was used to segment the background saliency map to obtain the foreground seed set and the foreground seed-based saliency map.(2)Introduce the center prior.The foreground-center map was obtained by fusing the foreground-seed-based saliency map.At the same time,a new fusion method is adopted,that is,the fusion of background and foreground saliency map to obtain the fusion saliency map.(3)Introduce the dark channel prior.Fusion step(2)obtains the saliency map and the dark channel prior saliency map,so as to obtain the rough saliency map.(4)Optimize the rough saliency map.In order to suppress the background noise in the rough saliency map,a weakening function is proposed.At the same time,the Gaussian model based on the center position of the salient object is given considering that the salient object is not in the center position of the image.In this paper,four publicly available saliency detection data sets are evaluated with 9other algorithms,and the evaluation results verify the reliability of the proposed algorithm.At the same time,the effect of parameters on the experimental results is verified and the validity of each stage of the algorithm is analyzed.