Object-Based Multi-feature Competitive Model for Visual Saliency Detection

Abstract

Saliency detection is an important topic for the researches on visual attention selective computation model. This paper aims at the conflict between the local saliency detection and the huge computational cost, and proposes an object-based multi-feature competitive model for visual saliency detection. The proposed model is based on the biologically motivated color-antagonistic model and is advanced by introducing the saturation channel. Gabor wavelet decomposition is used to obtain the feature maps. Object-based local inhibition is also proposed to accelerate the generation of feature conspicuity maps. The competition between multiple feature channels is simulated to implement the fusion of various visual features. The simulation results show that the proposed model has some advantages in local saliency detection and in computational speed.