Salient Region Detection From High Resolution Remote Sensing Image Based On Fusion Of Segment And Line Primitives

The visual attention mechanism is a preprocessing mechanism of the human visual system for visual information. Under its action, human eye can quickly notice the area of interest in a complex visual scene, while ignoring other information. The area attracting human attention is called saliency area. Simulating human visual attention mechanism by computer, and detecting the regions of the input image for visual saliency detection (hereinafter referred to as the saliency detection), we can distribute the priority of image processing according to the detection results in the subsequent image processing and target detection process. In this way, it can more rationally to allocate computer resources, and effectively improve the ability of the computer to understand the image.The saliency detection of natural scene image has already got rich research results. However, the research on remote sensing image is still an open topic. The reason is that there is a huge difference between the characteristics of remote sensing image and the natural scene image, and the more complicated background of remote sensing image leads to a less difference between foreground and background. So, the existing detection algorithm is difficult to play a role. With the appearance of high spatial resolution remote sensing image (high resolution remote sensing image), it is difficult to deal with the analysis of the remote sensing images. So, the application of visual saliency detection in high resolution remote sensing image processing has high research potential and application value.In this paper,based on the information extraction for remote sensing image especially high resolution remote sensing image, we study the method of detecting the visual salient region on remote sensing image. The main contents include:(1) The saliency detection algorithm based on linear feature is proposed. Firstly,we extract the linear information in the imagery, including the edge line and phase line. By analyzing the rule degree and density of lines in the image, we calculate the significant value of each region in the image. Finally, we propose a method of inverse distance transform for generating visual saliency map based line feature.(2) A saliency detection algorithm binding line features and patch features is proposed. Firstly, we make image into patches through image segmentation, then generate the visual saliency map by calculating the regional difference and overall differences of the patch's characteristics which is include brightness, spectral and degree rules of shape. Finally, we combine it with line-based feature visual saliency map through the global consolidation method, and get the final saliency map.(3) We compare the algorithm in this paper with other algorithms. The performance of the proposed algorithms and other commonly used algorithms is compared by the fixed threshold segmentation experiment and dynamic threshold segmentation to get results. Finally, we compare the results of experiment by visual analysis and quantitative evaluation.Experimental results show that the algorithms we propose have better detection results on visual saliency detection for the remote sensing images. And it also shows that our algorithms are more consistent with the characteristics of remote sensing images. It can accurately detect the salient area in a complex background and has good application value in the field of remote sensing.