Saliency detection based on global structure similarity metrics
LI Chong-fei · Computer Engineering and Science · 2013
Saliency detection is an important step image processing with computer, which includes the target recognition, image segmentation, and it is widely used in many fields such as network graphics and fingerprint recognition. In this paper, we proposed a global approach of structural similarity measure to significant target detection: using structural similarity to the human visual system’s high-level Abstraction, and taking into account the human visual characteristics and psychological feelings. We used the theory of structural similarity to measure the significant level of the target object, and get the saliency map after weight averaging the results. At last, the method of threshold was used to extract the salient target. Compared with the classic Itti algorithm, our approach can not only overcome the mosaic and take into account the global spatial information on the impact of structural similarity, but also select the parameters according to computing power and precision.