Research And Application On Image Saliency

The salient region of an image is the area where is attractive to visual attention most.Human visual system can quickly search the salient region of an image,But due to the computer only can store an image by pixel-wise,So an computer can hardly understand the image saliency.It need the help of relevant algorithm.The algorithm of image saliency detection is used to solve this problem.Image saliency is usually used as the pre-work about image processing.It was widely used by image segmentation,target tracing,self-adaption image editing as etc.The work of this article mainly focus on the image saliency algorithm.According to Histogram based Contrast(HC)and Region based Contrast(RC)algorithms,It made some improvement and proposed some application on the improvement.The main work are as follows:1.Introduced some classical saliency algorithms.These algorithms are proposed from different sides:methods based on Global Contrast,methods based on Region Contrast,methods based on image Statistics and Probability Theory,methods based on Frequency Transformation,etc.Analyzing on these algorithms by experiment we came to the conclusion that contrast based computing can result good performance.2.Based on HC and RC,This article proposed a new saliency detection algorithm: Superpixel segmentation based saliency detection algorithm.The algorithm first segment the image by SLIC SuperPixel segmentation.Then it extracts the salient feature and generates a saliency map.Compared with the HC and RC methods,the improve space are as follows:(1)Introducing super pixel segmentation algorithm to solve the RC background region in the defects of fixed size,by calculating a super pixel area and its surrounding contrast to extract salient areas of adaptive background region,improving the effect of the saliency.(2)When calculating the color features,instead of the HC method's color quantization and smooth,using SuperPixel area's Gaussian curve fitting the color histogram distribution to let the color features to get the maximum retention,at the same time,taking region and global into account when calculating the color contrast enables the saliency map more outstanding the whole salient target area.(3)First use gray level co-occurrence matrix to extract the texture feature combined with the color feature and spatial distributionto consist SuperPixel region feature,then multiply three components to generate the final saliency map.3.Based on the improved algorithm proposed in this article,introduces the application of saliency detection in image processing,it has been verified by experiment that salient keep on image enhancement method is superior to the traditional adaptive contrast method,the method of positioning on the target positioning accuracy is higher than that of other saliency detection method.