Depth-based saliency estimation for omnidirectional images

Visual saliency estimation aims at identifying and localizing the areas of images and videos that are attractive for a human subject. In this work, a novel approach for estimating the visual saliency in omnidirectional images is proposed. It is based on the identification of low-level image feature descriptors (i.e., the presence of texture, edges, etc.) coupled with the information about the local depth of the scene. To evaluate the performances of the proposed method, the estimated saliency map is compared with the available ground truth through two objective metrics: the correlation coefficient and the Kullback-Leibler divergence. The analysis of the achieved results confirms the validity of the proposed approach.