Influence updating and compactness enhanced approach for saliency detection
Zhihua Zhang, Zhengyi Liu · 2017
Salient object detection aims to detect the attractive objects to human viewers in an image. In order to solve the low detection precision problem of existing algorithms, a new method based on influence updating and compactness feature is proposed. Firstly, it propagates the background information to the entire image by using influence updating algorithm and obtains the initial saliency map. Then, it introduces the compactness feature, a low-level cue of image, to obtain the compactness of regions by computing the difference of color space and location space respectively. Finally, initial saliency map is enhanced by the compactness feature and pixel-level saliency correction. Experiment results on MSRA-1000, CSSD and ECSSD databases demonstrate that the proposed method performs well against the state-of-the-art methods by fusing prior knowledges and visual cues in terms of accuracy.