Image Salient Region Detection Based On Visual Attention

The salient target or region of an image refers to the object which attracts people's attention easily.Saliency detection aims to process or ignore image information selectively through using the human visual attention mechanism simulated by computer,and to get the salient target or region automatically.With the development of the image processing technology and computer science,image saliency detection has been widely used in machine learning,image processing and other fields.The research of image saliency detection has been become one of the hotspots in computer vision.Along with the intelligent development of the computer vision system and the increasing application areas,the image saliency detection technology has various drawbacks,although it has obtained a good research achievement.Therefore the saliency detection has a big progress space.This paper offers two methods by studying the image salient region detection technology based on human visual attention:1.A saliency detection method is proposed based on region contrast to extract the salient region of the image accurately.The method first segments the input image using superpixel segmentation and gets several regions to keep the shape of salient target.Then the region contrast is calculated by features of color and spatial and obtains the original saliency map to distinguish the salient target from background roughly.Furthermore,to improve the quality of saliency map,the method rises the region contrast to pixel contrast by combining with the contrast of average vectors of pixel features between image sub-region and its neighborhood.Finally,the proposed method assigns the higher value to the pixels which near the center,and generats the final saliency map according to the center prior.In order to evaluate the effect of the proposed algorithm,this paper has carried on the experiment to the 1000 natural images in MSRA database,and compared with other saliency detection methods.The experimental results show that the presented method can detect the salient region more exactly and suppress the background better.2.In consideration of the importance of priori knowledge to saliency detection,a saliency detection method is proposed based on the priori knowledge of boundary in the view of image boundary information.The image is first segmented into a number of superpixels by superpixel segmentation algorithm,which could remain the edges information.Then we put the superpixels as background in which the color is similar to the boundary by combining with the theory of boundary prior.Owing to the complexity of natural image,the four boundaries of image are different.In order to improve the accuracy,this paper processes the four boundaries respectively to get the saliency maps which are combined to generate the saliency map based on boundary prior.Since the different scales of superpixel effect the boundary selection,we processes the original image with different superpixel scales and fusions the results to obtain the final saliency map which extract the saliency region from background.Acrossing to a lot of experiments and the objective and subjective evaluations,the results show that this method can highlight the salient region and suppress the background effectively.