点群ベースの測定基準による埴輪の三次元顔類似性に関する研究

Abstract

The Haniwa are a type of terracotta figurines from ancient Japan, which have been scanned and saved as three-dimensional models composed of point cloud and triangular meshes for archaeological research. Although these three-dimensional models can be used to analyze the production technique of the Haniwa, there are several cases where archaeologists subjectively classify the Haniwa using only visual and qualitative observation approaches based on the Haniwa features, i.e., the face, clothes, etc. Here, the need for a computer-aided and quantitative evaluation method, using three-dimensionally measured point cloud, becomes evident. For a quantitative evaluation of the Haniwa, the evaluation of face similarity is considered an important issue. As the face is mainly composed of nose, mouth, and eyes, the face similarity of the Haniwa can be evaluated by comparing the corresponding facial parts between individuals. However, shapes and placements of the same facial parts for each pair of Haniwa have individual differences. Therefore, the face alignment to precisely localise the facial parts and recoginize their shapes should be required. To achieve the aforementioned precise alignment, a hybrid method combining ellipsoid-fitting-based segmentation and topological hole detection is proposed, to extract the point cloud of the nose and locate the contours of the eyes and mouth. Besides the face alignment, the quantitative evaluation essentially contains a process of comparing the faces by analyzing the relationships between the facial parts. For the face comparison, when the facial parts are aligned, the point clouds of the nose for each pair of Haniwa are registered through a scaling iterative closest point algorithm to achieve a value for scale alignment. After the point clouds of all the facial parts for each pair of Haniwa are aligned on the same scale to suppress scale error generated in production, their difference is calculated using the standard iterative closest point algorithm to obtain point-based dissimilarity for each pair of Haniwa. Moreover, owing to basic visual observation, the difference in the relative positions of the eyes and mouths with respect to those of the noses was manually measured as another dissimilarity for face comparison. Our experimental results demonstrate that the proposed method is effective for the similarity evaluation of the Haniwa through the three-dimensional point cloud.