Saliency Detection Based On Texture Inhibition And Continuous Distribution Estimation

With the rapid development of information technology, the size and complexityof the image data are increasing. Big data era puts forward new demands for theimage processing. As one of the hot spots in research of image area, saliency detectiontechnique has even become one of the most significant approaches in improving thefiltering and analysis accuracy of image data.Saliency detection technology based onvisual attention mechanism plays a important role in research and informationprocessing,the development of artificial intelligence, computer vision, and otherfields,.First of all, this paper introduces the basic theory of saliency detection, and thenanalyze the process of visual information transfer model and two kinds of visualattention mechanism. Finally this paper studies the classical Itti model and CA, GB,MZ, RC, SR five kinds of typical saliency detection method, and finish the simulationexperiment.As to the problem that the traditional methods of saliency detection tend to blurthe border and internal texture of interested target usually destroy the integrity of thetarget by applying the center-surround differences. This paper proposes a saliencydetection method based on texture inhibition and continuous distributeestimation.Firstly, bilateral filter is used to smooth texture disturbances of the targetand background region, while retain the major edges between the target andbackground.Then SLIC superpixel image segmentation algorithms is used to dividethe pixels into groups which have lots of pixels with the same characteristics. Thenfeatures of segment region are extracted by multidimensional normal distribution, andfeature saliency is calculated by two norm Wasserstein distance. Finally, the local andglobal saliency can be detected respectively by combining center-surround differenceswith background prior. The experiment result shows that the experiment method inthis paper can extract saliency region effectively. Finally,the full text is summarized and prospected.