Curvature-Based Feature Extraction Method for 3D Model Retrieval

Curvature on the surface of 3D mesh model is an important discrete differential geometrical descriptor. It can show the curving degree very well for those models with curving pieces and the ones with extreme points or extension components. In this paper, we use mean curvature and corresponding coordinates of the vertexes on the surface as the feature descriptor of model. The descriptor we defined describes models accurately and keeps invariant for transformation and rotation. Additionally, we bring EMD (Earth Moving Distance) method into the similarity measure frame. This comparison way is very efficient and especially adapt to these conditions: different numbers of the two features, variant or indefinite dimension of the features. Thought several retrieval experiments, the results implied that this descriptor could be used to find the exact model from the PSB model library. Particularly for models which are made up of curving pieces and those models with extreme points or parts, the result showed much better than other feature descriptors.