Research On Saliency Calculation Of Color Perception

Image saliency detection is designed to simulate the human visual system to analyze the image content to get the most noticeable part of the image.It is mainly used to handle the tasks of human eye fixation prediction and saliency object detection.Although the research on image saliency has achieved great success,there are still obvious shortcomings.Mainly reflected in following: first,the research on color and saliency does not strictly exclude the influence of other factors other than color on saliency,and the types of colors included are very limited;second,although there are relevant researches on color related saliency,due to the lack of targeted dataset,only the color factors independently affecting the saliency of the image have not received attention;third,the saliency detection model based on color images is not specifically consider the human eye's perception of color.In view of the above problems,this paper systematically studies the color saliency and the calculation of saliency based on color.The main content and innovations include:(1)The first eye-tracking dataset containing color images and gray images with the same content is constructed using an eye-tracker for research on color saliency.At the same time,we expand the research of existing significance detection models in color and gray-scale mode respectively,and prove that some saliency detection models get different detection results in color and gray-scale mode,so we suggest that the saliency detection model should be designed separately in color and gray-scale modes of the image.(2)Based on the color-aware saliency eye-tracking dataset,color saliency is analyzed in color spaces such RGB,lab and HSI.Firstly,according to the correlation between color and saliency in different color spaces,the Lab color space is determined to be the most suitable for studying color and saliency,and subsequent research is conducted in the Lab color space.Through further study of color and saliency in the Lab color space,some color effects are found: some specific colors attract more attention than other colors,and some specific color block combinations also attract more attention.(3)Based on the experimental conclusions about color and saliency,the saliency calculation of color image is modeled in this paper.A depth neural network is designed forthe prediction of human eye point and a saliency object detection model is improved.The proposed network for human eye fixation prediction outperforms the existing methods on three datasets;the improved salient object detection model integrates the perception characteristics of human eyes on color,and improves the extraction of content information substantially,which is conducive to the detection of salient objects closer to human visual attention in natural scenes.