Research On Image Salient Region Detection Based On Bottom-Up Model

Humans can quickly extract the most valuable information from complex natural scenes,and focus on the region of interest.Image salient region detection is a technology which can make computer simulate human's visual attention system,extract the region which is meet the condition of human attention mechanism from an image,simplify the follow-up computer process.The task of saliency detection is to estimate the position of salient object in an image.Specially,this detection is to input an image into the model and then to obtain a gray map,where the intensity of each pixel indicates the probability of being the salient region.This paper mainly study saliency detection based on Bottom-Up model.(1)In order to overcome the problems of imprecise in background extraction and weak ability of anti-background existed in traditional approaches of background priors,a salient object detection method using the global contrast and background priors is proposed.Firstly,a global-based saliency map is calculated by contrasting global color and the foreground seed is collected;Second,those superpixels with large difference are selected as the background seeds by comparing each boundary superpixels with the foreground seeds,improving the deficiency of the traditional algorithm which is directly from boundary background.Next,a new method of multi-interest Gaussian model is propsed to improve the sensitivity of the traditional center priori method to target location,which is more effective to fit salient target and reduce background interference.Compared with some traditional methods on publicly available benchmark(MSRA-1000),the simulation results demonstrate that the salient object detection approach proposed in this chapter performs more effective in dealing with complex boundary information and suppressing noise comparing with conventional methods.(2)In order to overcome the problems of nonuniform in salient object highlight and weak ability of anti-background existed in traditional approaches of saliency detection,a salient object detection method based on multi-scales contrast and bayesian model is proposed.Firstly,K-means algorithm based on traditional segmentation method is proposed,decreased the number of superpixels.Then,the convex hull center constraint based on multi-scale feature contrast method is used to obtain more uniformity foreground map.Secondly,use the Bayesian model to eliminate background noise.At the same time,in order to eliminate small background noise after the Bayesian model processed,the final salient object is obtained by multiplying foreground graph and salient graph.In compared with some traditional methods on publicly available datasets,the simulation results demonstrate this approach performs more uniform in highlight salient object,higher precision ratio and lower mean absolute error.(3)In order to deal with more complex and multi-objective scene,combination of multi-feature extraction and absorption Markov chain saliency detection method is proposed.Firstly,salient value re-sorting clustering method is used to extract more appropriate foreground and background absorption points.And then in order to combine the advantages of foreground and background maps,a new salient graph fusion method proposed which can increase the contrast of salient region and background.In compared with other methods on two publicly available datasets,the simulation results demonstrate that the salient object detection approach proposed in this chapter performs better and more effective.