Robust image saliency detection can process the image correctly without any prior knowledge and additional assumptions. Therefore, the saliency detection is still one of the important steps in the field of computer vision including object recognition and tracking, image and video encoding and image segmentation. Although infrared imaging has extensive applications, there is few saliency extraction algorithms based on infrared spectroscopy. We propose an infrared image-based saliency extraction algorithm based on human vision and information theory. The proposed algorithm uses both human visual attention mechanism and theory of information, and it can also produce a saliency image with full resolution. The detection results of the proposed algorithm get a higher accuracy and better recall rate, when tested on one of the largest infrared data sets which is publicly now and a data set created by ourselves.