Image Saliency Detection Algorithm Based On Visual Perception And Objectness

The development of machine vision technology let us enjoy the high-speed and convenient of information age,and the research of machine vision is based on the processing of images,Extracting useful information from the picture has become one of the key research topics in machine vision.Through the detection of saliency regions,we can effectively distinguish the foreground and background of the image,and get more attractive areas for human vision.Based on that,we can achieve more effective applications of machine vision,such as image segmentation,object recognition,image compression,and so on.In this paper,we introduce several classical image saliency detection model,and analyse the characteristics of each model,In view of this situation which existing models are not work very well,we learn from the models before,and propose a saliency detection algorithm which combined the Chromatism perception,Blurriness perception and Objectness of image.Human visual perception refers to two factors:chromatism perception and Blurriness perception.The chromatism is calculated in the Lab color space,The SLIC algorithm is used to divide the image into many superpixel blocks,construct a Markov random walk graph model.The background nodes and the foreground nodes are selected as the absorption state respectively,and the Markov absorption probability in the corresponding state is calculated.The similarity of the color features between the superpixel blocks is used as a measure of their significance.The Objectness is calculated on multiple scales,we combine the color contrast,the edge density and the superpixel span features to measure the probability of a pixel as an object component.At first,combining the chromatism and objectness,we propose a model to calculate the significant value of the image.With the further research,We find that the blurriness perception has more consistent with the selective attention mechanism of human vision.Therefore,we add this feature into the original model.We assume that focusing phenomenon exists in the edge pixels of the picture,and the radius of the circle of confusion is used to approximate the value of Blurriness.According to Weber-Fechner Law,We take logarithmic operation for human visual perception index,And add the influence of the objectness,Finally,we get the final saliency map in terms of probability.The image saliency detection algorithm has more analysis of the features of the image,and introduces human visual attention elements.Through qualitative and quantitative evaluation,It is proved that this model can detect salient regions more accurately.And this model can be utilized in the practical projects.