A multi feature fusion method for reassembly of 3D cultural heritage artifacts

Abstract The reassembly of cultural heritage (CH) artifacts from their fragments is an important problem in field archeology. We present a novel framework that is intended to fuse multiple local features that are the key principles utilized by archeologists. The framework extends the boundary contours to boundary bands, by defining the general adaptive neighborhood (GAN) of points on boundary contours, thus the necessary intermediate comparison between fragments is converted into the matching of GANs. Hence, we propose a novel local shape descriptor, oriented local alphabetic pattern (OLAP), to describe the local shape of the GANs. Then, three different strategies for GANs matching are introduced. Finally, the initial position is calculated according to the sets of appropriate points on boundary contours, and the pairwise alignment is performed by the iterative closest point method (ICP). This framework is effective in composing thin-shell and thick-shell fragments, especially those containing flat or incomplete fracture regions whose pairwise matching is usually unreliable and ambiguous, hence their reassembly remains challenging to existing algorithms. Experimental results with real point clouds are presented to demonstrate the efficiency and superiority of our framework on different datasets.