In the last years, visual saliency has become a challenging research field, and a big number of computational models were developed. While detecting salient object in still images was well studied, video saliency detection is in the early stages. In this paper, we propose a novel video saliency detection method based on Boolean maps. Unlike still images, video frames are characterized by statistic and dynamic information. A set of Boolean maps are generated by thresholding feature channels (color and motion features). Using the gestalt principle for figure-ground segregation, saliency prediction is derived from the Boolean maps where connected regions are marked as salient. Our proposed method is evaluated over two video saliency benchmark datasets and compared to seven state-of-the-art methods. Results have shown that our method outperforms other methods on the two datasets.
展开▼
