Automatic 3D model reconstruction of the diamond object

Abstract

This paper addresses the problems presented by automatic 3D modeling and measurement in evaluating the diamond-cutting grade of the round brilliant cut diamond. (In this paper, we refer to the round brilliant cut diamond, the most common diamond, unless otherwise specified.) Diamond features such as small size (most are less than 1cm), lack of texture, highlights, and light transmission intensify these problems, and commonly used methods in computer vision such as laser triangulation, structured light, stereo matching, and shape from shading fail to generate the diamond’s 3D model. Based on the image features of the diamond and the polishing process of the rough diamond, this paper proposes a “shape from multi-view closed contours” (SFMCC) method. First, the diamond is placed on a turntable, and a single CCD camera with one telecentric lens captures silhouette images from multiple views as the turntable rotates. Next, accurate closed contours are extracted from the silhouette images sequence, and a 3D model of the diamond can be retrieved through the space-carving process of the bounding box with multi-view closed contours. To implement the SFMCC method, a complete system including hardware and software is designed and constructed. Many different experiments have been designed to test this method, and results have proven its high efficiency and accuracy.