Matching in the Dark: A Dataset for Matching Image Pairs of Low-light Scenes

This paper considers matching images of low-light scenes, aiming to widen the\nfrontier of SfM and visual SLAM applications. Recent image sensors can record\nthe brightness of scenes with more than eight-bit precision, available in their\nRAW-format image. We are interested in making full use of such high-precision\ninformation to match extremely low-light scene images that conventional methods\ncannot handle. For extreme low-light scenes, even if some of their brightness\ninformation exists in the RAW format images' low bits, the standard raw image\nprocessing on cameras fails to utilize them properly. As was recently shown by\nChen et al., CNNs can learn to produce images with a natural appearance from\nsuch RAW-format images. To consider if and how well we can utilize such\ninformation stored in RAW-format images for image matching, we have created a\nnew dataset named MID (matching in the dark). Using it, we experimentally\nevaluated combinations of eight image-enhancing methods and eleven image\nmatching methods consisting of classical/neural local descriptors and\nclassical/neural initial point-matching methods. The results show the advantage\nof using the RAW-format images and the strengths and weaknesses of the above\ncomponent methods. They also imply there is room for further research.\n