Saliency Calculation Based on the Fusion of Enhanced Contour Features and Spatial Semantic Information

Aiming at the inability of the models for salient object detection to accurately separate the salient foreground and background under complex backgrounds,and important semantic information is not accurately detected by the low-level visual model,a salient feature model that combines enhanced contour features and spatial semantic information is proposed in this paper to improve the results of saliency detection.Firstly,the region modularization is completed by the regional gradient smoothing method,and each region of image is calculated using the texture compactness to obtain salient regions,which reduces the calculation amount of subsequent network training.Then,an enhanced contour network based on local regions is proposed for prediction.The salient maps at each branch are fused,and a spatial semantic feature network is proposed based on visual perception to mine the deep details of the image.Finally,the unified network framework is used to integrate the enhanced contour features and spatial semantic features to obtain the final fine distinctive features,and more complete and clear contours are retained.The algorithm of this paper is tested with other popular algorithms in the HKU-IS,ECSSD and DUT-OMRON databases.And the recall rate and average error of the proposed algorithm are better than other algorithms.