In this paper, we propose a data index structure which is constructed by a small autonomous mobile robot so that it manages millions of subimages it takes during a navigation of dozens of minutes. The subimages are managed according to a similarity measure between a pair of subimages, which is based on a method for quantizing HSV colors. The data index structure has been inspired by the CF tree of BIRCH, which is an early work in data squashing, though care and inventions were necessary as the bins of HSV colors are highly correlated. We also propose an application for peculiar subimage detection by the robot, which exploits the data index structures for the current image and another one for all images in its navigation. Experiments conducted in a private office of about 25m 2 proved the feasibility of the data index structure and the effectiveness of the peculiar subimage detection.