Salient Object Detection Based On Eye Tracking Data

Saliency detection for image has become a key step for computer vision applications,which has been widely used to numerous vision problems,such as image retrieval,target recognition and image compression etc.Saliency detection model aims to detect the region of interest in images and provides an efficient solution for image semantic understanding.Eye tracking technology can accurately tracks human eye movement and reflects the visual perception.In this paper,we propose a saliency detection model based on eye tracking.Eye tracking data can improve the performance of saliency detection model and the relevant to the human visual perception.The main works are summarized as follows:1)An eye tracking database of MSRA-1000 has been built.We collect eye tracking data from ten observers.The eye tracking database lay the foundation for the follow-up study.2)We present a new superpixels segmentation based on eye movement saliency value.The simple linear iterative clustering(SLIC)algorithm is optimized by eye movement saliency value.The experiment results show that our method can effectively reduce the complexity.3)A novel saliency detection model base on eye tracking is proposed.We use a new selection strategy of training samples based on eye tracking.The saliency detection model is constructed by multi kernel boosting learning,and optimized by the saliency map based on statistical and multi-scale strategy.The experimental results on MSRA and DUT-OMRON databases show that our method perform favorably against 10 state-of-the-art saliency methods in terms of P-R curve,AUC and F-measure.