Visual Saliency Detection Based On Multiple-feature

Saliency detection can effectively separate the attention and non-attention regions ofthe image. Through obtain images of the area of interesting, we can carry out imageenhancement, feature extraction, image edge detection, image recognition andunderstanding, image cutting processing. As the basic subject of computer vision, it hasimportant research significance and value. There is an urgent need to design more accurateand robust saliency algorithm after the sequence of image processing and research.According to the reference model saliency detection can be divided into biologicaldetection method, pure computational method, and the combination of these two saliencydetection method. Through analysis the different image feature, and combined withimproved k-means algorithm, and propose a new computational saliency detection modelbased on multiple feature. Firstly, in order to obtain better clustering, the classification ofthe image is obtained by clustering, not only consider the color of the pixel but also thespatial distance between pixels. The initializing center according to the honeycombconjecture select a regular hexagon instead of a regular rectangular. Secondly, to be able toget the full resolution image, and highlight the whole target, it needs to analysis theimage’s color, texture and spatial distance feature according to the local contrast andglobal contrast, generate eight feature maps. Finally, we through effective integration toget the final saliency map.Through implement some experiments on Achanta’s data set with manually labeledground truth. This paper compares the result from two aspects of subjective evaluation andobjective evaluation experiment contrast with some the-art-of-state methods. Theexperimental results demonstrate the effective of the proposed method.