Abstract

Feature descriptors have become a ubiquitous tool in shape analysis. Features can be extracted and subsequently used to design discriminative signatures for solving a variety of 3D shape analysis problems. In particular, shape classification and retrieval are intriguing and challenging problems that lie at the crossroads of computer vision, geometry processing, machine learning and medical imaging.

In this thesis, we propose spectral graph wavelet approaches for the classification and retrieval of deformable 3D shapes. First, we review the recent shape descriptors based on the spectral decomposition of the Laplace-Beltrami operator, which provides a rich set of eigenbases that are invariant to intrinsic isometries. We then provide a detailed overview of spectral graph wavelets. In an effort to capture both local and global characteristics of a 3D shape, we propose a three-step feature description framework. Local descriptors are first extracted via the spectral graph wavelet transform having the Mexican hat wavelet as a generating kernel. Then, mid-level features are obtained by embedding local descriptors into the visual vocabulary space using the soft-assignment coding step of the bag-of-features model. A global descriptor is subsequently constructed by aggregating mid-level features weighted by a geodesic exponential kernel, resulting in a matrix representation that describes the frequency of appearance of nearby codewords in the vocabulary. In order to analyze the performance of the proposed algorithms on 3D shape classification, support vector machines and deep belief networks are applied to mid-level features. To assess the performance of the proposed approach for nonrigid 3D shape retrieval, we compare the global descriptor of a query to the global descriptors of the rest of shapes in the dataset using a dissimilarity measure and find the closest shape. Experimental results on three standard 3D shape benchmarks demonstrate the effectiveness of the proposed classification and retrieval approaches in comparison with state-of-the-art methods.