A Learning-Based Positive Feedback Approach in Salient Object Detection

Motivated by human visual system, a HELM-based positive feedback approach is proposed which intensifies salient objects and promotes the saturation of human perception on a fixation area. Randomly-weighted deep neural networks are applied online to our method and lead to stable performance. Our algorithm is data-driven totally and does not need any prior knowledge compared to the existing approaches. Experimental results based on the standard image datasets show that our approach is an effective one to enhance the performance compared to the existing algorithms.