Computational model for salient object detection with anisotropy

Abstract

An innovative computational model for salient objects detection is proposed. The model is based on the exploitation of the anisotropy property of images by means of pixelwise directional entropy. The generalized Rényi entropy and the discrete cosine transform (DCT) coefficients are selected for this purpose. An entropy map of an input image can be obtained by calculating the Rényi entropy via local patch-based DCT. Analyzing the statistical property of the power spectrum of the entropy map on log-log scale, we find the power law is also appropriate for entropy maps. Accordingly, a saliency map can be derived from the entropy residual computation. Salient objects are detected using a seeded region growing algorithm. Both qualitative and quantitative experiments are conducted. The corresponding results demonstrate the outstanding performance of the proposed model.