Automatic stitching of panoramas for geological images of polished sections

Abstract. This paper addresses the challenge of constructing panoramic images from a set of overlapping microscopic images of geological polished sections.This task is valuable both for automatic mineral detection and for the creating of advanced virtual educational collections. We present LumenStone P1, a new dataset of polished section images for panorama stitching, along with an automatic method for constructing panoramas of these images. Comparisons with existing approaches, including commercial ones, demonstrate that our method is comparable in stitching quality. The numerical value of panorama stitching quality is 0.33 by RMSE for our dataset. The results demonstrate the effectiveness of the proposed method and its practical applicability for the preparation of series of images and further computer analysis of panoramic geological images of polished sections. The LumenStone P1 dataset intended for panorama creation, as well as the source code of the proposed method, are publicly available.