Research on Video Saliency Detection Via Contrast and Self-Adaptive Transfer

Although a lot of studies in salient motion detection have achieved great success in recent years, many challenges still exist toward the video saliency detection over the non-stationary videos and videos with slowly-moving objects, which supposes to exhibit significant influence on its corresponding subsequent applications. Thus, it urgently needs a more robust, stable, and precise method to solve the above mentioned limitations. In fact, inspired from the basic visualization rule of the human vision system, the human’s attention can be easily attracted by two independent factors: the motion saliency clue and the color saliency clue. Hence, this paper develops a novel salient motion detection method by fusing the motion saliency with the color saliency, which refines the preliminary saliency map by self-adaptive transfer via the newly designed intra-frame correlation. Also, comprehensive experimental results of our method toward the state-of-the-art methods over 4 public available benchmarks demonstrate the superiority of our method both in its robustness and high detection precision.