Research On Image Saliency Detection Algorithm Based On Local Background Priority Strategy

Image saliency detection is an important component of image recognition in intelligent processing,which can be traced back to 1998,Itti et al.The results show that the human visual system depends on the global features in the process of scene recognition,and the recognition of the object depends on the local features.By studying many reference materials,the advantages and disadvantages of some detection algorithms have been summarized.In order to improve the applicability of the saliency algorithm and to ensure the integrity of the results,a new algorithm of image saliency detection based on local background priority strategy is proposed.This paper starts with the background and present situation of the topic.The significance of the research on the saliency model has been discussed.A brief introduction to the human visual system has been given.The descriptive features commonly used in image saliency detection have been introduced.The work of this paper mainly includes the following three aspects:Color features selection have been improved in this paper.RGB and Lab color space have been selected.The adjacent color difference sequences of each color channel(6channels)have been calculated in two color space.The maximum of the color difference in the channel has been extracted selectively.The final sequence which is used to compared with the merged threshold has been generated.In this paper,the selection of color feature is used in the local change of object edge,and the effective information is preserved to improve the performance of the algorithm.Scale adaptive threshold segmentation.A new matrix is generated by reading the static image in this algorithm.All the pixels in the static image and every two adjacent(4neighborhood)difference between pixels are included in the matrix.The original static image average difference is used as the threshold to sign whether all the differences in the new matrix are connected.Segmentation is achieved by connecting the results to the original pixel.Some restrictions are made in the segmentation block area according to the human visual identification scale.Adaptive threshold is generated according to the different block size by the inverse function model.And then comparison and merge are achieved.This part of the algorithm is extended to the conceptual object from the independent properties of the pixel blocks.This part of work is indispensable for subsequent research.Local background priority is analyzed and experimentally investigated in this paper.According to the relative position of the block of the line data extracted from the local area(background-subject-background),the background position is marked,and then the edge is optimized.Compared with other algorithms,image saliency detection algorithm based on local background priority strategy we proposed in this paper does not need to be connected with other threshold algorithm to generate two-value image.Adaptive threshold merging can eliminate the details of objects in complex environments and can focus on the saliency comparison of the same scale.The experimental results on a publicly available experimental data set show that the proposed algorithm is effective.