A robust approach for salient region detection

Salient region detection deals with finding those regions in an image which gains most of viewer´s attention. The various applications of saliency map include robotics, computer vision and data transmission. In this paper, we propose a method based on image contrast which uses poisson distribution. In the proposed approach, we have utilized the fact that if a pixel belongs to salient region then the difference between pixel intensity and mean will be more and if pixel is part of background then this difference will be less. No prior training is required in proposed approach. The proposed approach is evaluated on MSRA dataset and is compared with one of the image contrast based approach and the performance is consistent in terms of recall, precision and execution time.