Single Image Depth Prediction on One Perspective Cultural Heritage Artefact

Abstract

Archaeological archives are essential assets because they provide information as resources for research, mainly digital archives that are not degraded over time. Instead of directly visiting a site, an archaeologist can examine and manipulate the data without harming the real object. However, choosing an efficient scanning scheme with based on the Archaeological needs is a challenging task. First, we collect all available acquisition methods and compare each other. Based on the results, Single image Reconstruction is proved to be the most efficient. The central aspect of Single Image Reconstruction (SIR) is Single Image Depth Prediction (SIDP). However, the color and lighting problems arise and become a problem in predicting the depth. Thus, a robust depth prediction system for color and light problems is needed. Several steps are created in this thesis to cover the problems. An investigation about the most efficient method for archaeologists to scan a single perspective artifact is undertaken. This research is to determine which method performs efficiently regarding the requirement of an archaeologist. On the results, SIR is proved as a suitable method to obtain the detail on 3D surfaces. However, there are two problems that need to be solved to improve the results: differences in color material and differences in lighting. These problems affect the outcome of SIDP, one of the steps in SIDP. Furthermore, this thesis focuses on improving single image depth prediction to solve these two problems. The steps are: creating a new dataset, solving the color material problem, investigating a new approach for the lighting difference problem, and proposing a new method to solve material color and lighting difference problems, simultaneously.