Co-saliency Detection Algorithm Based On Cluster And Region Feature

The nature of image co-saliency detection aims to extract the common interesting saliency from the multiple images. This technology can be used in co-segmentation, image distance measurement, and weakly supervised learning, it has been a new research topic in recent years. However, with the rapid development of computer network and digital image processing technology, the number and resolution of the image increase rapidly, existing algorithms of images co-saliency detection cannot meet the actual demand on detection speed and quality, based on the problem about raising the speed and result, this paper proposed some improved algorithm as follows.Firstly, some algorithms of image co-saliency detection disrupted by the background noise easily, and it leads the generating saliency maps would be in the wake of serious background noise, in addition, these algorithms has some other problems, such as large computation and long detection time. This paper applies prior foreground to remove partial noise at first, and then give a co-saliency detection using contrast cue and spatial cue, compared with traditional methods, this algorithm not only overcome some of the background interference to generate co-saliency maps with lower noise, but also the calculation speed has been improved in recent times.Secondly, co-saliency detection includes two important parts, i.e. saliency detection and collaboration detection, the detection result of each part would directly affect the overall effect, Considering that some algorithms ignore the result of single-image saliency map, this paper use single image saliency detection based on Markov chain combining collaboration detection for images pair based on Simrank similarity algorithm, the background of those saliency maps which are obtained from this algorithm have been effectively suppressed, and enhance the brightness of foreground. Proved that through the experiment, the result of this co-saliency detection method has been raised by big percentages.At last, this paper applies co-saliency detection method to person re-identification. The saliency detection is used for recognizing the target area of human images, and the global distribution of co-saliency detection is used for measuring the similarity of target area. The experiment proves the effectiveness of person re-identification based on co-saliency detection method.