Dejan V. Vranic

Feature-based similarity search in 3D object databases

The development of effective content-based multimedia search systems is an important research issue due to the growing amount of digital audio-visual information. In the case of images and video, the growth of digital data has been observed since the introduction of 2D capture devices. A similar development is expected for 3D data as acquisition and dissemination technology of 3D models is constantly improving. 3D objects are becoming an important type of multimedia data with many promising application possibilities. Defining the aspects that constitute the similarity among 3D objects and designing algorithms that implement such similarity definitions is a difficult problem. Over the last few years, a strong interest in methods for 3D similarity search has arisen, and a growing number of competing algorithms for content-based retrieval of 3D objects have been proposed. We survey feature-based methods for 3D retrieval, and we propose a taxonomy for these methods. We also present experimental results, comparing the effectiveness of some of the surveyed methods.

Tools for 3D-object retrieval: Karhunen-Loeve transform and spherical harmonics

We present tools for 3D object retrieval in which a model, a polygonal mesh, serves as a query and similar objects are retrieved from a collection of 3D objects. Algorithms proceed first by a normalization step (pose estimation) in which models are transformed into a canonical coordinate frame. Second, feature vectors are extracted and compared with those derived from normalized models in the search space. Using a metric in the feature vector space nearest neighbors are computed and ranked. Objects thus retrieved are displayed for inspection, selection, and processing. For the pose estimation we introduce a modified Karhunen-Loeve transform that takes into account not only vertices or polygon centroids from the 3D models but all points in the polygons of the objects. Some feature vectors can be regarded as samples of functions on the 2-sphere. We use Fourier expansions of these functions as uniform representations allowing embedded multi-resolution feature vectors. Our implementation demonstrates and visualizes these tools.

3D Model Retrieval with Spherical Harmonics and Moments

We consider 3D object retrieval in which a polygonal mesh serves as a queryand similar objects are retrieved from a collection of 3D objects. Algorithms proceed first bya normalization step in which models are transformed into canonical coordinates. Second, feature vectors are extracted and compared with those derived from normalized models in the search space. In the feature vector space nearest neighbors are computed and ranked. Retrieved objects are displayed for inspection, selection, and processing. Our feature vectors are based on rays cast from the center of mass of the object. For each raythe object extent in the raydirection yields a sample of a function on the sphere. We compared two kinds of representations of this function, namelyspherical harmonics and moments. Our empirical comparison using precision-recall diagrams for retrieval results in a data base of 3D models showed that the method using spherical harmonics performed better.

DESIRE: a composite 3D-shape descriptor

The topic of this communication is shape-similarity search for 3D-mesh models. We present and evaluate a composite 3D-shape feature vector (DESIRE), which is formed using depth buffer images, silhouettes, and ray-extents of a polygonal mesh. We contrast our method with the approach that is declared the best in the recent study. Our experiments suggest that the composite feature vector, which is extracted in a canonical coordinate frame, generally outperforms the competing method, which relies upon pairwise alignment of models. We also provide a Web-based retrieval system as well as publicly available executables for verifying the results.

Description of 3D-shape using a complex function on the sphere

We propose a novel feature vector suitable for searching collections of 3D-object images by shape similarity. In this search, a polygonal mesh model serves as a query. For each model, feature vectors are automatically extracted and stored. Shape similarity between 3D-objects in the search space is determined by finding and ranking nearest neighbors in the feature vector space. Ranked objects are retrieved for inspection, selection, and processing. The feature vector is obtained by forming a complex function on the sphere. Afterwards, we apply the fast Fourier transform (FFT) on the sphere and obtain Fourier coefficients for spherical harmonics. The absolute values of the coefficients form the feature vector. Retrieval efficiency of the new approach is evaluated by constructing precision/recall diagrams and using two different 3D-model databases. We compared the approach with two methods based on real functions on the sphere. Our empirical comparison showed that the complex feature vector performed best. We also prepared a Web-based retrieval system for testing the methods discussed.

An experimental effectiveness comparison of methods for 3D similarity search

Methods for content-based similarity search are fundamental for managing large multimedia repositories, as they make it possible to conduct queries for similar content, and to organize the repositories into classes of similar objects. 3D objects are an important type of multimedia data with many promising application possibilities. Defining the aspects that constitute the similarity among 3D objects, and designing algorithms that implement such similarity definitions is a difficult problem. Over the last few years, a strong interest in 3D similarity search has arisen, and a growing number of competing algorithms for the retrieval of 3D objects have been proposed. The contributions of this paper are to survey a body of recently proposed methods for 3D similarity search, to organize them along a descriptor extraction process model, and to present an extensive experimental effectiveness and efficiency evaluation of these methods, using several 3D databases.

Automatic selection and combination of descriptors for effective 3D similarity search

We focus on improving the effectiveness of similarity search in 3D object repositories from a system-oriented perspective. Motivated by an effectiveness evaluation of several individual 3D retrieval methods, we research a selection heuristic, called purity, for choosing retrieval methods based on query-dependent characteristics. We show that the purity selection method significantly improves the search effectiveness compared to the best single methods. We then show that retrieval effectiveness can be further boosted by considering combinations of multiple retrieval methods to perform the search. We propose to use a dynamically weighted combination of feature vectors based on the purity concept, and we experimentally show that the search effectiveness of our combined methods by far exceeds the effectiveness of our best implemented single method.

Using entropy impurity for improved 3D object similarity search

Similarity search in 3D object databases is becoming an important problem in multimedia retrieval, with many practical applications. We investigate methods for improving the effectiveness in a retrieval system that implements multiple feature extraction algorithms to choose from. Our techniques are based on the entropy impurity measure, widely used in the context of decision trees. We propose a method for the a priori estimation of individual feature vector performance, given a query. We then define two approaches that use this estimator to improve the retrieval effectiveness. Our experimental results show that significant improvements are achievable using these methods

An experimental comparison of feature-based 3D retrieval methods

3D objects are an important type of multimedia data with many promising application possibilities. Defining the aspects that constitute the similarity among 3D objects, and designing algorithms that implement such similarity definitions is a difficult problem. Over the last few years, a strong interest in methods for feature-based 3D similarity search has arisen, and a growing number of competing algorithms for content-based retrieval of 3D objects have been proposed. We present an extensive experimental evaluation of the retrieval effectiveness and efficiency of a large part of the current state-of-the-art feature-based methods for 3D similarity search, giving a contrasting assessment of the different approaches.

Content-based search for 3D-objects

The topic of this paper is content-based 3D-object retrieval. The approach is based on feature vectors, which capture 3D-shape of a model represented as a triangle mesh. The feature vectors are invariant with respect to translation, rotation, scaling, and reflection and robust with respect to level-of-detail. Before the feature extraction, each 3D-object is transformed (normalized) into a canonical position and orientation. The search is performed in the feature vector space in which the feature vector of a query model is used as a key. Original normalization steps and feature vectors are presented in this communication.