Alessandro Pianezzi

Semester Work
Supervisors: Janine Thoma, Dr. Ajad Chhatkuli, Prof. Dr. Luc van Gool

Deep Learning for Tooth Outline Detection

In order to improve the handling of intra-oral dental scanners, Thoma et al. [1] have developed an AR application, which is capable of blending the 3D computer model into the user's field of view in real time. In this semester project, we investigate the possibility of detecting tooth boundaries in order to improve localization of the 3D overlay. For this purpose, we first label tooth boundary pixels in 50 images. We use our dataset to train a fully convolutional neural network to detect tooth boundaries and classify them into different kinds of tooth boundaries. We conduct experiments and show through quantitative analysis, that the tooth boundary detector has potential to be incorporated into the AR intra-oral scanning framework. [1] J. Thoma, M. Havlena, S. Stalder and L. V. Gool, "[POSTER] Augmented Reality for User-Friendly Intra-Oral Scanning," 2017 IEEE International Symposium on Mixed and Augmented Reality (ISMAR-Adjunct), Nantes, 2017, pp. 97-102.