3D-model retrieval using bag-of-salient local spectrums (BOS)

The interest of this paper is to present a new efficient method for indexing and retrieval of three-dimensional objects in general databases. The proposed method is based on Bag of Features (BoF) paradigm, and salient characteristics of objects. We first extract the salient points of target object using a scalar function calculated on the surface. Then, each interest point is associated with local Fourier descriptor. This descriptor is computed in its neighborhood by projecting the geometry onto the first eigenvectors of Laplace-Beltrami operator. Additionally, in an offline learning stage, a visual dictionary is built by clustering a large set of feature descriptors. Then each 3D shape is described by a histogram of occurrences of these visual words weighted by his local descriptors number. The obtained results show the highly discriminative capability of the proposed approach against rigid and non-rigid transformations, noise and geometry changes. The performance of our approach is also demonstrated for global and partial shape retrieval.