Visual Saliency Estimation through Manifold Learning

 Most early work makes more effort to build saliency models on low-level image features based on local contrast. These methods investigate the rarity of image regions with respect to (small) local neighborhoods.  Recent efforts have been made toward global contrast based saliency estimation, where saliency of an image region is evaluated at the global scale with respect to the entire image.  From previous research it is known that producing a correct salient map is sensitive to the size of regions, and a manual fine tuning of the segmentation is a prerequisite for some challenging images.  In summary, it has been observed that most global approaches depend on either regions from image segmentation, or blocks with specified sizes.  While image structures at global scale are usually an important factor for producing salient stimuli, this paper presents a novel approach for saliency modeling( that too manifold based saliency estimation)