Learning partial differential equations for saliency detection

Abstract

Learning-based partial differential equations (PDEs), which combine fundamental differential invariants into a nonlinear regressor, have been successfully applied to several computer vision and image processing problems. However, it cannot apply to saliency detection directly. In this paper, we present a novel learning-based PDEs model and learn the PDEs from training samples. We simplify the current model by setting the indicate function constant along with the evolution process. When learned the PDEs, we first combine three simple priors for the pre-processing and then solve the PDEs to generate the final saliency map for each image. Experimental results on public benchmark data set MSRA-1000 demonstrate the superiority of our hybrid approach.