Research On Image Saliency Object Detection Algorithm

The saliency object detection of images is an important application in current machine vision work,and the effective detection effect brings great convenience to computer vision processing.As an image preprocessing process,it can detect the region of interest in the image and extract important information of the image to reduce the amount of calculation.This paper mainly studies the image saliency object detection from three aspects: seed point information diffusion,pattern mining and extreme learning machine.Firstly,saliency object detection on fusion foreground and background seeds diffusion algorithm is proposed in the paper.It is very important to generate effective seed points for effective saliency maps.In this paper,the foreground seeds are taken into account in the selection of background seeds,so that the seed point information is more abundant and comprehensive;Next,the background seeds and the foreground seeds are respectively diffused,and their respective saliency maps are fused;Finally,the saliency map is improved through clustering optimization and suppression functions,so that the salient area is more prominent and the background area is more suppressed.Secondly,In order to find more accurate seed point information,this paper proposes a significant object detection algorithm based on pattern mining based on the association rules of data.Firstly,the pattern mining is used to find the more accurate foreground points in the image;Then the feature similarity of the foreground points and the remaining nodes in the image are measured,and the initial saliency map is obtained;Finally,the structural information diffusion optimization is used to obtain the final saliency map.The pattern mining has found the foreground point very well,so the algorithm has clear object and high recognition rate.Finally,In order to improve the accuracy of detection,a multi-scale saliency object detection algorithm based on extreme learning is proposed,inspired by the task of data binary classification.First,In the super-pixel segmentation of a single scale,the CNN feature training extreme learning machine is used to obtain the label of the imagesuper-pixel,and the initial saliency map is calculated by the label result;then the Gaussian function and the smooth prior are used to optimize the saliency map;The saliency maps at the three scales take the mean of the superimposed saliency maps as the final saliency map,so that the effect of the salient map is more perfect.