A study on interest point guided visual saliency

Visual attention is one of the most critical characteristics of human visual system (HVS), which infers the attractive regions in a visual scene. It has been an active research topic over the past decades and many proposed models of visual attention have demonstrated successful applications in a wide range of fields including computer vision and image processing. On the other hand, interest point detection is another hot topic that leads practical contributions to the real-time applications such as visual retrieval and augmented reality. In this paper, we try to investigate the relationship between the interest point and the visual attention. An informative analysis is reported by comparing the performance of different interest point models in predicting the visual fixation. It is found that the blob based interest point model generally outperforms the corner based model. Furthermore, we propose a mixture strategy by integrating all the interest point algorithms, and the experimental results indicate that this proposed method is competitive with some state-of-the-art algorithms.