A New Hybrid Approach for Saliency Detection in Infrared Images

Saliency detection in infrared images plays a critical role in large amounts of practical applications, such as infrared image compression, target detection and tracking. A novel saliency detection method in a single infrared image is proposed in this paper. First, a local sparse representation based approach is designed to calculate the initial saliency map for an input infrared image. Then, to further remove the background information in the initial saliency map, a novel method based on two-dimensional maximum entropy/minimum cross entropy and maximum standard deviation is proposed to predict the foreground. By subtracting the predicted foreground from the original infrared image, the background information can be obtained. Finally, the initial saliency map is refined through the background information. The presented method is evaluated on the real-life infrared images and the experimental results show that the proposed method achieves better performance compared to the state-of-the-art algorithms.