Image Saliency Target Detection Based On Prior Knowledge And Propagation Mechanism

In the era of big data,the explosion of data leads to more and more informat ion need to be dealt with.To efficiently and accurately extract the critical information from a massive collection of image data has become more important in computer vision field,which brings the research on image saliency detection.Reliable saliency detection as a pre-processing can contribute to many computer vision applications,such as image segmentation,image categorization,target recognition and image retrieval.However,the researchers are faced sundry challenges in their detection,such as the multiple target,the complex scene of the image and so forth,making it still a challenging task to construct a reliable detection algorithm.Therefore,it is of great significance to carry out in-depth research on the topic of saliency detection.The main research of this paper is based on the bottom-down model.Through the study of the existing methods at domestic and overseas,two kinds of significant detection methods are put forward.Firstly,we propose a saliency detection method based on hybrid pri or knowledge and cellular automata.The results of the existing algorithms using a single prior knowledge can not well highlight the object.For the shortcomings of the method using single prior knowledge,we make the coarse map can better highlight the sa lient object by a hybrid prior knowledge;then,a novel propagation method based on cellular automata is introduced to optimize the results.The experimental results show that the model can obtain reliable and accurate results.Secondly,we propose a new saliency detection method based on boundary prior and propagation mechanism.Existing methods to extract the background directly using four boundaries are not good enough to deal with scene that the target appears at the boundary.For the shortcoming of this method,a reliable method for extracting background is presented by removing the superpixels in the border set if they have large difference.Then,step-wise propagation mechanism is proposed to calculate the saliency and improve the accuracy.The first step extracts extract background via robust background information,which can provide foreground information for second step.The next step detect the salient regions through foreground cues.The experimental results show that this method can effectively h ighlight the target uniformly and suppress the non-salient regions.We validate our algorithm using several public databases and experimental results demonstrate that the proposed model performs better than the state-of-the-art methods,which synchronously pop out object and suppress background.