Thermal–visible registration of human silhouettes: A similarity measure performance evaluation

When dealing with the registration of information from different image sources, the de facto similarity measure used is Mutual Information (MI). Although MI gives good performance in many image registration applications, recent works in thermal–visible registration have shown that other similarity measures can give results that are as accurate, if not more than MI. Furthermore, some of these measures also have the advantage of being calculated independently from each image to register, which allows them to be integrated more easily in energy minimization frameworks. In this article, we investigate the accuracy of similarity measures for thermal–visible image registration of human silhouettes, including MI, Sum of Squared Differences (SSD), Normalized Cross-Correlation (NCC), Histograms of Oriented Gradients (HOG), Local Self-Similarity (LSS), Scale-Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), Census, Fast Retina Keypoint (FREAK), and Binary Robust Independent Elementary Feature (BRIEF). We tested the various similarity measures in dense stereo matching tasks over 25,000 windows to have statistically significant results. To do so, we created a new dataset in which one to five humans are walking in a scene in various depth planes. Results show that even if MI is a very strong performer, particularly for large regions of interest (ROI), LSS gives better accuracies when ROI are small or segmented into small fragments because of its ability to capture shape. The other tested similarity measures did not give consistently accurate results.