Detect saliency to understand a photo

Visual saliency is an extremely hot topic in image and video processing. Our paper addresses a novel method for saliency detection, which conveys the photographers´ idea of a scene. It consists of analyzing the photo elements like color, depth-of-field (DOF) and composition. We use the color spatial distribution feature to calculate the initial saliency. A simple but efficient classifier is trained to discriminate the low DOF image. For this kind of images we find the focal area to refine the initial saliency. The saliency map will be further adjusted by an image composition analysis step to generate the final output. We compare our framework with state-of-the-art algorithms. Experimental results show that, our method out-performs the previous methods in many cases, especially in the low DOF situation.