Saliency Object Detection Research Of Multi-information Fusion

In the rapid development of computer vision research field,saliency object detection can not only improve the accuracy of image detection,but also further accelerate the speed of image detection.Aiming at the existing main problem of texture detail information not obvious and the edge contour information incomplete in the current saliency object detection algorithms,saliency object detection research algorithm of multi-information fusion is proposed.The algorithm mainly includes three processes,namely pre-processing process,multi-information extraction process and fusion optimization process.Firstly,in the pre-processing process,the points near the boundary are removed by the edge operators to obtain the pre-processing maps.Then,in the multi-information extraction process,use the proposed convex-hull contrast method,scale optimization color feature method and unsharp mask optimization LBP operator method to extract contrast feature,color feature and texture feature respectively,and obtain contrast feature maps,color feature maps and texture feature maps.Use th e best algorithm in center prior algorithms and edge prior algorithms respectively to obtain center prior maps and edge prior maps.Finally,in the fusion optimization process,the primary sa liency maps are obtained by using the multi-layer cellular automata fusion 3 feature maps and 2 prior maps,and then the primary saliency maps are optimized by the fast guided filter to obtain the final saliency maps.On the 3 public MSRA10 K,ASD and ECSSD image datasets,the paper proposed algorithm and 12 classical algorithms conduct subjective comparison experiments and objective comparison experiments.The experimental results show that the paper proposed algorithm satisfies the visual experience of the human eye,as the same time it can accurately and comprehensively extract saliency object regions,and retain more texture information and complete edge information,the evaluation indicators are better than the comparison algorithms,have good reliability and adaptability.This paper has 37 figures,2 tables and 67 references.