Visual attention is one of the most important mechanism of HVS (human visual system) and has been applied into many fields. Research on visual attention model is hot and difficult. This paper presents a novel visual attention model for video based on NMFSCP (non-negative matrix factorization sparseness on parts). Saliency map of this model is generated by utilizing four types of visual attention features such as intensity, color, orientation and motion. Motion feature of video key frame is extracted by the NMFCP (Non-negative matrix factorization sparseness on parts) algorithm. Intensity, color and orientation features were obtained by the Itti visual attention model. Four features are combined with unequal linear coefficients according to the ratio of motion block in video frame. Simulation result shows the efficiency of proposed model.
展开▼