Salient Object Detection Based On Machine Learning

Human eyes can find distinctive objects or regions quickly and accurately in complex scenes for differentiation,which is the visual attention mechanism of human eyes.Salient object detection is based on the visual attention mechanism,aiming to use the visual attention mechanism of computer simulation of human eyes to find the most salient objects in mages.As a visual task preprocessing process,in other computer vision,pattern recognition task has played an important role,such as image segmentation,image compression,visual tracking,scene reconstruction and so on.Now,Salient object detection is mainly divided into two parts:single image salient object detection and co-salient object detection.Single image salient object detection aims to detect significant targets in a single image,while co-salient object detection aims to detect common salient objects in a set of images.With the wide application of Depth information,Depth information has been proved to be a very practical feature in the field of salient object detection.Therefore,the salient object dection are also gradually extended from RGB image to RGB-D image.Therefore,single image salient object detection can be divided into RGB image salient object detection and RGB-D image salient object detection,while co-salient object detection can be divided into RGB co-salient object detection and RGB-D co-salient object detection.This paper mainly focuses on RGB co-salient object detection and rgb-d salient object detection.In the existing RGB co-salient image salient object detection,a group of images is not really considered as a whole,which makes it difficult to obtain good co-salient consistency and contains many co-salient backgrounds.At present,there are two problems to be solved in RGB-D salient object detection.One is to design Depth feature;the other is to effectively combine RGB feature and Depth feature.Traditional model by manually design features,although can achieve a certain effect,but the degree of generalization is very low,and based on RGB characteristics and characteristics of the Depth processing,respectively,and then the fusion of two streams of significant figure to get the final salient map,although can get certain achievements,but the Depth map if the effect is very poor or completely useless will have great influence on the final result.In this paper,we propose a novel framework to obtain the consistency of co-salient objects,and can reduce the interference to the co-salient objects in the complex background.The model includes a cluster-based tree-structured sparsity-including regularization that make region from same class have identical saliency value,and a Laplacian constraint regularization is also integrated into the model,the purpose is to smooth the salient value in same cluster.Furthermore,to facilitate the efficient,a coherence weight is identified and integrated into the model.Through matrix decomposition,the original feature space is decomposed into low-rank part and sparse part,and the sparse part is the final co-salient objects.The RGB image co-salient object detection method proposed in this paper is compared with other state-of-the-art methods in the three public datasets of iCoseg,iCoseg-sub,Image Pair.The experimental results show that our method has better results.For RGB-D salient object detection,so salient object detection method by Single Stream Recurrent Convolutional Neural Network(SSRCNN),in which color feature is served as primary cues and depth feature is served as auxiliary ones,is proposed.First RGBD four-channels input is fed into VGG-16 net to generate multiple level features which express the most original feature for RGB-D image.The coarse saliency map from the deepest features can detect and localize salient objects,but loss the boundaries and subtle structures.So Depth Recurrent Convolution Neural Network(DRCNN)is then applied to each level feature for rendering salient object outline from deep to shallow hierarchically and progressively.With the help of deeper level feature,original depth cue and coarse saliency map,each level feature can accurately predict the salient objects in different scales.At last all the saliency maps from each level are fused together to generate final results.In this paper,the RGB-D single-stream network salient object method proposed in this paper and other state-of-the-art image salient object detection methods are verified and evaluated on the four most widely used NLPR1000 and NJU2000 RGB-D datasets.The experimental results show that the method proposed in this paper has achieved good results.In this paper,the above two proposed for RGB co-salient object detection and RGB-D salient object detection,and good results are obtained,which provide a variety of ideas for the research of salient object detection and prepare for the research work in the field of computer vision and pattern recognition.