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

3D objects are an important multimedia data type with many applications in domains such as Computer Aided Design, Simulation, Visualization, and Entertainment. Advancements in production, acquisition, and dissemination technology contribute to growing repositories of 3D objects. Consequently, there is a demand for advanced searching and indexing techniques to make effective and efficient use of such large repositories. Methods for automatically extracting descriptors from 3D objects are a key approach to this end. In this paper, we survey techniques for searching for similar content in databases of 3D objects. We address the basic concepts for extraction of 3D object descriptors which in turn can be used for searching and indexing. We sketch the wealth of different descriptors by two recently proposed schemes, and discuss methods for benchmarking the qualitative performance of 3D retrieval systems.

An important research issue in multimedia databases is the retrieval of similar objects. For most applications in multimedia databases, an exact search is not meaningful. Thus, much effort has been devoted to develop efficient and effective similarity search techniques. A recent approach, that has been shown to improve the effectiveness of similarity search in multimedia databases, resorts to the usage of combinations of metrics where the desirable contribution (weight) of each metric is chosen at query time. This paper presents the Multi-Metric M-tree (M 3-tree), a metric access method that supports similarity queries with dynamic combinations of metric functions. The M 3-tree, an extension of the Mtree, stores partial distances to better estimate the weighed distances between routing/ground entries and each query, where a single distance function is used to build the whole index. An experimental evaluation shows that the M 3-tree may be as efficient as having multiple M-trees (one for each combination of metrics).

We present an extension of the fuzzy c-Means algorithm, which operates simultaneously on different feature spaces—so-called parallel universes—and also incorporates noise detection. The method assigns membership values of patterns to different universes, which are then adopted throughout the training. This leads to better clustering results since patterns not contributing to clustering in a universe are (completely or partially) ignored. The method also uses an auxiliary universe to capture patterns that do not contribute to any of the clusters in the real universes and therefore likely represent noise. The outcome of the algorithm are clusters distributed over different parallel universes, each modeling a particular, potentially overlapping, subset of the data and a set of patterns detected as noise. One potential target application of the proposed method is biological data analysis where different descriptors for molecules are available but none of them by itself shows global satisfactory prediction results.

In this paper, we propose and evaluate a novel shape measurement describing the extent to which a 3D mesh is rectilinear. Since the rectilinearity measure corresponds proportionally to the ratio of the sum of three orthogonal projected areas and the surface area of the mesh, it has the following desirable properties: 1) the estimated rectilinearity is always a number from (0,1]; 2) the estimated rectilinearity is 1 if and only if the measured 3D shape is rectilinear; 3) there are shapes whose estimated rectilinearity is arbitrarily close to 0; 4) the measurement is invariant under scale, rotation, and translation; 5) the 3D objects can be either open or closed meshes, and we can also deal with poor quality meshes; 6) the measurement is insensitive to noise and stable under small topology errors; and 7) a Genetic Algorithm (GA) can be applied to calculate the approximate rectilinearity efficiently. We have also implemented two experiments of its applications. The first experiment shows that, in some cases, the calculation of rectilinearity provides a better tool for registering the pose of 3D meshes compared to PCA. The second experiment demonstrates that the combination of this measurement and other shape descriptors can significantly improve 3D shape retrieval performance.

We systematically evaluate a recently proposed method for unsupervised discrimination power analysis for feature selection and optimization in multimedia applications. A series of experiments using real and synthetic benchmark data is conducted, the results of which indicate the suitability of the method for unsupervised feature selection and optimization. We present an approach for generating synthetic feature spaces of varying discrimination power, modeling main characteristics from real world feature vector extractors. A simple, yet powerful visualization is used to communicate the results of the automatic analysis to the user.

Methods for management and analysis of non-standard data often rely on the so-called feature vector approach. The technique describes complex data instances by vectors of characteristic numeric values which allow to index the data and to calculate similarity scores between the data elements. Thereby, feature vectors often are a key ingredient to intelligent data analysis algorithms including instances of clustering, classification, and similarity search algorithms. However, identification of appropriate feature vectors for a given database of a given data type is a challenging task. Determining good feature vector extractors usually involves benchmarks relying on supervised information, which makes it an expensive and data dependent process. In this paper, we address the feature selection problem by a novel approach based on analysis of certain feature space images. We develop two image-based analysis techniques for the automatic discrimination power analysis of feature spaces. We evaluate the techniques on a comprehensive feature selection benchmark, demonstrating the effectiveness of our analysis and its potential toward automatically addressing the feature selection problem.

Similarity searching in metric spaces has a vast number of applications in several fields like multimedia databases, text retrieval, computational biology, and pattern recognition. In this context, one of the most important similarity queries is the k nearest neighbor (k-NN) search. The standard best-first k-NN algorithm uses a lower bound on the distance to prune objects during the search. Although optimal in several aspects, the disadvantage of this method is that its space requirements for the priority queue that stores unprocessed clusters can be linear in the database size. Most of the optimizations used in spatial access methods (for example, pruning using MinMaxDist) cannot be applied in metric spaces, due to the lack of geometric properties. We propose a new k-NN algorithm that uses distance estimators, aiming to reduce the storage requirements of the search algorithm. The method stays optimal, yet it can significantly prune the priority queue without altering the output of the query. Experimental results with synthetic and real datasets confirm the reduction in storage space of our proposed algorithm, showing savings of up to 80 % of the original space requirement.

In recent years, there has been enormous growth in the number of 3D models and their availability to a wide segment of the population. Examples include the National Design Repository which stores 3D computer-aided design (CAD) models for tens of thousands of mechanical parts, the Protein Data Bank (PDB) that has atomic positions for tens of thousands of protein molecules, and the Princeton Shape Benchmark with thousands of everyday objects represented as polygonal surface models. With the availability of free interactive tools for creating 3D models and graphics cards for home computers, we can expect 3D data to become ever more widely available. Given the availability of 3D data, searching for a 3D object in a large database is a core problem for numerous applications including object recognition and the reuse of expertly created data. This raises two key research problems: 1) How can we improve search techniques? and 2) How do we evaluate 3D search techniques? The first contribution of this dissertation is an analysis technique to select the most important or distinctive regions of an object. Our approach identifies regions of a surface

Visualization of 2D point clouds is one of the most basic yet one of the most important problems in many visual data analysis tasks. Point clouds arise in many contexts including scatter plot analysis, or the visualization of high-dimensional or geo-spatial data. Typical analysis tasks in point cloud data include assessing the overall structure and distribution of the data, assessing spatial relationships between data elements, and identification of clusters and outliers. Standard point-based visualization methods do not scale well with respect to the data set size. Specifically, as the number of data points and data classes increases, the display quickly gets crowded, making it difficult to effectively analyze the point clouds. We propose to abstract large sets of point clouds to compact shapes, facilitating the scalability of point cloud visualization with respect to data set size. We introduce a novel algorithm for constructing compact shapes that enclose all members of a given point cloud, providing good perceptional properties and supporting visual analysis of large data sets of many overlapping point clouds. We apply the algorithm in two different applications, demonstrating the effectiveness of the technique for large point cloud data. We also present an evaluation of key shape metrics, showing the efficiency of the solution as compared to standard approaches.