Saliency Object Detection Algorithm Based On Globaland Local Fusing

Vision is an important channel for human beings and other intelligent creatures to know the world.The data shows that 80% of the world's information is transmitted through the visual.The amount of information acquired by human vision is very large.With the help of the visual system,human can quickly and accurately locate the area of interested or target from a complex scene.With the progress of society and the arrival of the era of big data,digital image has gradually become an indispensable information carrier in people's daily life.But there are a lot of redundant data in the image.It is the focus of this paper that how to introduce the visual neural system into the computer,so that the computer can quickly and accurately extract the salient region of the image.Firstly,an improved method for the detection of the boundary is proposed.After studying the pre-paper of geodesic saliency using background priors algorithm,we found that directly use the prior background can quickly obtain a super pixel and the boundary space weight relations,and then draw the saliency map.But even if there is only a small part of the significant object is connected to the border,it will also directly defined as the background.So if there is a significant object part which connects with the boundary in the image boundary,the saliency object contour will have a great error.In order to improve the algorithm of the global detection method based on the boundary,our algorithm add the color contrast,and the whole color region as a part to draw a significant figure,through experiments,we found that this algorithm is very good to solve the defect of the original algorithm.Then,a new visual saliency algorithm based on global and local fusion is proposed.By comparing the advantages and disadvantages of the global detection algorithm and the local detection algorithm,we find that fusing both of them can achieve better results,but the existing fusion algorithms are simple additive or multiplicative,which is simple and not robust.In this paper,a new algorithm based on global and local algorithms is proposed,which combines the foreground and background of local clustering segmentation algorithm and the global description algorithm based on boundary.It also be made a special optimization at the boundary of the object boundary to high light the salienct object.The final result is proved to have some advantages compared with other algorithms.