Saliency fusion via sparse and double low rank decomposition.

•A novel sparse and double low rank decomposition model (SDLRD) is proposed.•We develop a SDLRD-based saliency fusion framework.•An effective ADMM-based optimization algorithm is proposed to solve the SDLRD model.•We cast the saliency fusion as a subspace decomposition problem.•Experiments on five datasets show the effectiveness of the proposed method.