3D refined reconstruction for brushes based on multiple images

Abstract

Chinese Calligraphy is one of the most esteemed visual art all over the world, and ink brushes are the traditional writing tool for Chinese calligraphy. We want to combine the traditional culture with the latest computer vision techniques, and an exact 3D model of brushes is a necessary prerequisite. However, traditional ink brushes have elongated and thin bodies, which is tough to create dense point clouds and an exact 3D reconstruction model. In this paper, we represent a new approach to achieve a complete 3D reconstruction model of brushes within an acceptable time. Our approach utilizes special brushes and extracts interesting points by using both SIFT and AKAZE. Through the combination of two feature extraction algorithms, interesting points could be doubled and be more focused on brushes rather than backgrounds. Moreover, we compare computational costs in different parameters in our 3D reconstruction pipeline and summary some experience to accelerate the reconstruction process. We decrease the number of neighbor cameras when creating a depth map, and reduce the number of output points to the extent that naked eyes cannot recognize the differences. Experimental results show that our approach is able to create a complete and consistent 3D reconstruction model within three minutes.