Research On Image Saliency Detection Algorithm Combining Color And Texture Features

Images can simply and quickly convey a variety of information,and they are also the most commonly using information carriers in people's daily life.By the description of the image makes things more intuitive and concise,so that some scenes which are difficult to describe in language can also be expressed through images.When more and more things need to be expressed with images,more redundant information will be mixed into it,and it is difficult to obtain important information.To solve this problem,researchers in the field of computer vision have proposed the concept of image saliency detection,which is one of the attention mechanisms that can select important regions from the picture.The selected regions contain the main information of the picture,which can be used to extract intermediate and higher-level information and prepare for the next image processing.Image saliency detection greatly improves the efficiency of image processing and saves the utilization of resources.In recent years,there have been a lot of researches on saliency algorithms based on graph models.These algorithms use the relationship between image regions to obtain saliency maps,which has the advantages of low computational complexity and good robustness.This paper mainly studies the saliency of images and improves manifold ranking algorithms based on graph models,which include the selection of refined foreground and background labels,improving the original graph models and manifold ranking formula,and based on color features to add texture features.The main work and innovations of this paper are as follows:(1)In order to solve the problem of inaccurate label selection in manifold ranking algorithms,saliency detection algorithm based on manifold ranking and refined seed labels was proposed.Manifold ranking algorithms based on graph model need to build graph model and rely heavily on labels.When the selected labels are inaccurate,that will seriously affect the finally result.In this paper,the k-means clustering algorithms is used to optimize the background labels and the foreground labels,and the traditional graph model is improved.The improved algorithm can effectively remove redundant information,which is more easily to distinguish boundaries and emphasizes saliency regions uniformly.(2)In order to detect good saliency object when the image color contrast is low,this paper proposes fusing color and texture based on label weighting manifold ranking saliency detection algorithm.The improved algorithm uses the complementarity of color and texture to construct saliency map,then fuses the color saliency map and the texture saliency map.Based on the traditional manifold ranking algorithms,the labels in the improved algorithm are weighted by the degree matrix,then the labels that are more likely to be saliency regions having more contribute to the final saliency map.The improved algorithm can better detect salient object in complex scenes,even the inaccurate labels will not have a large impact on the final saliency map.The experimental results show that the saliency detection algorithm based on manifold ranking and refined seed labels has good visual effects,which can not only effectively suppress the background regions but also obtain a uniform overall target;Fusing color and texture based on label weighting manifold ranking saliency detection algorithm,which can obtain more complete saliency object and has good robustness.