Saliency Detection And Segment Based On Human Visual Attention Mechanism

Recently years, with the continuous development of the computer technology,intelligent devices are widely used in people’s daily life. According to statistics, sixtypercent information what people get that come from the vision, and image is thefundamental unit of visual information. So, computer vision technology is more important.Saliency detection studied in this paper is simulating human vision model to complete theimage segmentation and saliency region extraction. It can construct flexible and adaptivevision algorithms. The research can be applied to find lesion location positioning ofmedical image, detect industrial parts, track and recognize motion target, also thealgorithms would be used in image coding and compression, image matching andsynthesis, image retrieval, and auto-focus of camera and other fields. So, saliencydetection is attracted more and more attention of researchers.Our paper proposes a new saliency detection and segmentation algorithm based onsaliency map. The algorithm consists of three parts: global background difference mapgeneration, saliency map calculation and saliency segmentation and extraction. First,generate difference map by blurring original image to eliminate fine texture details as wellas noise and coding artifacts based on lab color space. And then get the saliency map byClustering image based on Climbing and k-means algorithm. Finally, extract saliencyregion as follows：1) analysis the saliency map;2) remove isolated salient points,3)enhance, extract and connect edges;4)mark connected domain.We use the salient object images which are created by Microsoft Research Lab of Asiato test our approach and compare with other five main approaches in the state-of-art, theexperimental result shows that our approach is better than others in precision, recall andF-Measure.