Structural Segmentation of Sherd Surfaces

Motivation

The goal of this project was to partition a surface into structurally meaningful components. For example, the surface structure of a broken piece of pottery may be partitioned into smooth inner and outer surface regions which are distinct from the irregular broken edge regions. In the field of archaeology, where pieces of broken pottery are called sherds, this shape and structural information is useful for classifying and eventually reconstructing broken vessels. We looked at the problem of structural segmentation of surfaces in this context. A similar study of surface structure is seen in computer graphics research. Issues of surface partitioning arise as a fundamental problem of surface representation. Partitions are used to guide a variety of geometric modeling operations, such as decimation and deformation.

Surface Structure of Sherds

The break surface of a sherd is defined as the surface that is exposed when a pottery vessel breaks. Generally, the break surface of a sherd is a long and narrow strip as compared to the overall sherd surface. Break surfaces may be used to measure how well two sherds fit together.

The inner and outer surfaces of a sherd refer to the smooth regions of the sherd's surface, which are named to correspond with the inside/outside relationship of the vessel as a whole. By nature of vessel construction on a potter's wheel these smooth inside and outside regions are surfaces of revolution, and hence contain valuable information regarding the vessel's profile and diameter. Estimates of these parameters may be computed from the segmented inner and outer surfaces.

To segment a sherd surface into structurally meaningful components (break surfaces vs. smooth regions), we have adapted a method for 2D image segmentation and applied it to the triangle meshes surface representation of the sherds.

Smooth

Smoother

Smoothest

Break Curves

The surface structure segmentation has improved. A surface segmentation is labeled with red and blue colours in the left figure. The break curves which separate red and blue are shown on the right.

Segmented collection.

And estimated break curves.