Similarity-based approach to earthenware reconstruction

This paper proposes an earthenware reconstruction system, which can automatically reconstruct some earthenware from numerous given potsherds in two-dimensional grayscale images. The system supposes that given potsherds are thin and moderately flat and small. Some earthenware having three-dimensional shape, such as crocks, can be reconstructed by the system within the supposition. The system performs earthenware reconstruction through two phases. At the first phase, potsherds are joined automatically in two dimensions. An efficient joint detection algorithm using surface pattern and shape similarity is proposed at this phase. At the second phase, three-dimensional shape is recovered by an adequate three-dimensional transformation. Some experimental results of reconstruction from numerous potsherds are also reported.