Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
530650 | Pattern Recognition | 2010 | 16 Pages |
This paper presents a new 3D shape representation and classification methodology developed for use in craniofacial dysmorphology studies. The methodology computes low-level features at each point of a 3D mesh representation, aggregates the features into histograms over mesh neighborhoods, learns the characteristics of salient point histograms for each particular application, and represents the points in a 2D spatial map based on a longitude–latitude transformation. Experimental results on the medical classification tasks show that our methodology achieves higher classification accuracy compared to medical experts and existing state-of-the-art 3D descriptors. Additional experimental results highlight the strength and advantage of the flexible framework that allows the methodology to generalize from specific medical classification tasks to general 3D object classification tasks.