Image Salient Object Detection Based On Visual Perception And Attention

With the popularity of digital camera and smart phone,as well as the rapid development of social networks and twitter,images are very rich in resources and convey a lot of message in human daily life.Whether all the content in an image is necessary for further image processing,saliency detection is worth investigating.Human can rapidly and accurately identify the salient region in scenes.When watching images or videos,not every pixel is of equal importance to us.In most cases,our visual attention mainly focuses on a salient sub-region of an image,and we follow this salient region in an image sequence.For simulating such capability of the human vision attention system in computer vision,saliency detection is generally considered as a selection process of the interesting regions that attract the visual attention in a scene.This capability has long been studied by cognitive scientists and has recently attracted a lot of interest in the computer vision community mainly because it helps find the objects or regions that efficiently represent a scene and thus harness complex vision problems.While the progress on accurate detection of the attentional regions for subsequent image processing,saliency detection still lack the generality and robustness inherent in the primate visual system.Inspired by the visual perception and attention,four novel saliency detection methods are proposed.These methods can accurately highlight the whole salient object and efficient suppress the background.The main contributions can be summarized as follows:(1)To preserve the locality and similarity of regions,we introduce a new self-representation learning for saliency region detection algorithm which explicitly considers the locality and the similarity among regions.A new saliency map is calculated by the self-representation learning,which explicitly explores the local spatial structure of salient objects and thus lo-cate more accurate salient regions.Moreover,we advance a background based dictionary from the border set of the image,by which objects can be more accurately located.Experi-mental results on six large benchmark databases demonstrate the proposed method performs favorably against the state-of-the-art methods in terms of five evaluation criterions.Mean-while,our method runs as fast as most existing algorithms.(2)Subspace representation based salient object detection has received increasing interests in recent years.However,due to the independent coding process of sparse reconstruction,the locality and the similarity among regions to be encoded are not explored.To preserve the locality and similarity of regions,a graph Laplacian regularization term is constructed as a smooth operator to alleviate the instability of the salient score in visual object.Then a new saliency map is calculated by incorporating this local graph regularizer into sparse re-construction,which explicitly explores the local spatial structure of salient objects and thus obtains more uniform salient map.Moreover,we advance a heuristic object based dictionary from background superpixels,by which objects can be more accurately located.Experimen-tal results on six large benchmark databases demonstrate that the proposed method performs favorably against fifteen state-of-the-art methods in terms of five evaluation criterions.(3)Recently,saliency detection has been becoming a popular topic in computer vision.An object-level saliency detection algorithm which explicitly explores bottom-up visual atten-tion and objectness cues is proposed..Firstly,some category-independent object candidates are segmented from the image by the quantized color attributes.Then two metrics,global cues and candidate objectness,are developed to estimate the saliency in the whole image and the object candidates respectively.We use global cues to describe the focusness and spatial distribution of color attributes in the entire image.As supplementary,candidate objectness can reveal the objectness of the object candidates.Finally,we integrate the two salien-cy cues to derive a saliency map of the image.By explicitly fusing candidate objectness and global cues,our proposed method is more suitable for processing images with com-plex background.Experimental results on three large benchmark databases demonstrate that the proposed method achieve more favorable performance than fifteen recent state-of-the-art methods in terms of five evaluation criterions.(4)Existing saliency detection methods are difficulty to highlight the whole salient region since the object is not always in the boundary of the image.Inspired by the human percep-tion,the semi-supervised smoothing representation is proposed to detect the salient object.First,a fixation prediction model is used to coarsely select the current region of interest in scenes that attracts visual attention and the initial attended view can be obtained.The bound-ary region without eye fixations can be called as background seeds.We rank the similarity of the image regions with foreground seeds or background seeds via semi-supervised smooth-ing representation.The final saliency map is built by normalizing and fusing foreground and background saliency maps of all seeds from both fixation and background regions.Exten-sive experiments on six public datasets demonstrate that the proposed method outperforms state-of-the-art methods.