Cross-Domain Co-Occurring Feature for Visible-Infrared Image Matching

As the two most commonly used imaging devices, infrared sensor, and visible sensor play a vital and essential role in the field of heterogeneous image matching. Therefore, visible-infrared image matching which aims to search images across them has important application and theoretical significance. However, due to the vastly different imaging principles, how to accurately match between visible and infrared image remains a challenge. In fact, the two images describe one scene from different aspects. There is a symbiotic relationship between their features, which we named as cross-domain co-occurring feature. In this paper, based on cross-domain co-occurring feature, we present a novel visible-infrared image matching algorithm. Concretely, co-occurring feature is first constructed by cross-domain image database and feature extraction approach. Then three visual vocabulary trees can be built by visible feature, infrared feature, and co-occurring feature. Thus, the symbiotic relationship between the two domains is established by cooccurring feature and vocabulary trees. With this relationship, each image is represented by a list of leaf node of co-occurring vocabulary tree. Finally, we measure the image similarity and the highest scoring image is the matching result. As a bi-directional method, we evaluate the proposed algorithm on two tasks: visible-to infrared matching and infrared-to-visible matching. Experiments on the Korea Advanced Institute of Science and Technology all-day place recognition database captured from 42-km sequences demonstrate that cooccurring feature is effectiveness and efficiency to link different domains. And the matching approach also achieves superior performance.