Saliency Detection Based On Imporved Position Prior

In the field of computer vision,saliency detection is a concerned research direction.Saliency map of image can be obtained by saliency detection algorithm.Saliency map annotates the saliency value of each pixel in the original image,and the saliency value represents the saliency degree of corresponding pixel.Correlation algorithm can quickly identify the salient region of image through saliency map,so saliency map can be applied to image Compression,scene classification,image retrieval and many other fields are used to reduce the computational complexity of related algorithms.Traditional saliency detection algorithms used to obtain saliency map by calculating the global contrast or local contrast of the image.These algorithms often fuse with the background prior knowledge or the foreground prior knowledge of the image.In most cases,the salient region is located in the center of the image,and the background area is located at the edge of the image.Therefore,according to the pixel's physical position in the image,some algorithms first default the pixels in the fixed area of the image as background or foreground,and then calculate the contrast between the whole image pixel and the region to obtain the saliency map,which is the traditional central prior method and background prior method.This prior method is more targeted than the traditional contrast method,but when the prediction of salient regions is not on time,the efficiency of the algorithm will be greatly reduced.Therefore,this paper proposes an improved prior method,which divides the foreground region and background region more accurately in the preprocessing stage.Firstly,a proper number of subgraphs are obtained by sliding window,and then the saliency of each subgraph is evaluated by ng vector.Then,the score of each sub image is transformed into the super pixel level fraction,and then the foreground node and background node are obtained by adaptive threshold segmentation.After obtaining high-quality foreground nodes and background nodes,the contrast calculation method is used to obtain high-quality saliency map.However,the saliency images obtained by the two algorithms are defective but complementary to each other.Therefore,this paper proposes an algorithm to fuse and optimize the two saliency maps,and finally obtain high-quality saliency maps.Finally,the proposed algorithm is compared with many classical saliency detection algorithms on SED2,THUR-10000 and ECSSD data sets.The experiments show that the proposed saliency detection algorithm is better than the classical algorithm in commonevaluation indicators,which proves the effectiveness of the proposed algorithm.