An experimental comparison of feature-based 3D retrieval methods

Arxiv preprint arXiv:1111.6387, 2011

Classifying 3D models into classes is an important step in 3D model retrieval process. However, most classification system does not include semantic information. In this paper, a new method has been proposed to classify and to retrieve 3D model using semantic concepts and ontology. First, we use the machine learning methods to label 3D models by k-means algorithm in shape indexes space. Second, classification of 3D models into classes based on semantic annotation is applied. Finally, ontology is constructed with instances for each class, which contain spatial relationships, shape indexes and measures. Therefore, the 3D model retrieval system is comprised of lowlevel and high-level visual features. We interpret our results using the Princeton Shape Benchmark Database and our prototypical 3D Model Search System (3DMSS). The results show the performance of the proposed new approach to classify and to retrieve 3D models.

In computer graphics, Content-Based Retrieval (CBR) is a database system that consists of receiving an object as input and returning a list of similar objects. Although originally conceived for images, CBR systems can work with objects of another nature such as sounds, and three-dimensional models , represented as geometrical meshes, transporting CBR systems to contexts other than images To assess the proximity between the meshes, some techniques consist of performing mathematical transformations of the meshes into feature vectors and calculating the distance relationship between these vectors. In this work, we built a prototype of a 3D CBR system that uses an order relation, instead of a distance function, as a similarity function between the objects. Furthermore , we propose the use of an order in R n that we call Extended Lexicographical Order (ELO), which seeks to take into account all the information present in the vectors to be compared. We compared our prototype not only with...

2006

This paper presents an approach to measure 3D similarity by combining two feature vectors. We extract the feature vectors by employing two similarity models: Direction Vector of Surfaces (DVS) and Shape Histogram of Projected Volume (SHV). Then we merge the features by two approaches: merging the two original feature vectors and merging computed-distances. Our experiments show that combining two features using either feature merging or distance merging enhances the retrieval performance. Furthermore, we show that employing weighting factor to the merging process implies differently to the retrieval performance, depending on data set distribution. Finally, we introduce an idea of meta feature-vectors which regards the already calculated distances as new feature vectors. Using this approach, a new similarity space might be established, and new distances could be calculated in order to enhance the performance.

In this papеr, we presеnt a new 3D modеl retriеval systеm basеd on the statistical selectеd featurе lеarning approach. This approach can be integratеd with any еxisting and potеntial 3D modеl retriеval algorithm which includеs 3D modеl featurе еxtraction, selеction and distancе computation. By applying the analysis of variancе of еach componеnt among all featurе vеctors and rеmoving thosе componеnts with F-valuеs smallеr than a critical F-valuе, we can reducе the featurе dimеnsion and also keеp featurеs with highеr discriminating ability. Moreovеr, the SVM classifiеr is learnеd basеd on the ANOVA-selectеd featurеs, which are extractеd from the training set. We conduct experimеnts using the McGill Articulatеd Shapе Bеnchmark databasе [1] for 3D modеl classification and retriеval, and demonstratе a significant performancе improvemеnt in the prеcision-rеcall curvеs.

Computer-Aided Design and Applications, 2008

In this paper, we present a comparative study that concerns relevance feedback (RF) algorithms in the context of content-based 3D object retrieval. In this study, we employ RF algorithms which range from query modification and multiple queries to one-class support vector machines (SVM). Furthermore, we employ pseudo relevance feedback (PRF) and show that it can considerably improve the performance of content-based retrieval. Our comparative study is based upon extensive experiments that take into account datasets containing generic as well as CAD models.

International Journal of Computer Science and Information Technology, 2011

The recent technological progress in acquisition, modeling and processing of 3D data leads to the proliferation of a large number of 3D objects databases. Consequently, the techniques used for content based 3D retrieval has become necessary. In this paper, we introduce a new method for 3D objects recognition and retrieval by using a set of binary images CLI (Characteristic level images). We propose a 3D indexing and search approach based on the similarity between characteristic level images using Hu moments for it indexing. To measure the similarity between 3D objects we compute the Hausdorff distance between a vectors descriptor. The performance of this new approach is evaluated at set of 3D object of well known database, is NTU (National Taiwan University) database.

2007

The increase in availability and use of digital three-dimensional (3D) synthetic or scanned objects, makes necessary the availability of basic database operations, such as retrieval. Retrieval methods are based on the extraction of a compact shape descriptor; the challenge is to design a shape descriptor which describes the original object in sufficient detail to make accurate 3D object retrieval possible. Building on previous work, this paper proposes a novel depth buffer-based shape descriptor (called PTK) which encompasses symmetry, eigenvalue-related weighting and an object thickness related measure to provide accuracy that surpasses previous state-of-the-art methods. Evaluation of the novel method's parameters as well as a direct comparison to other approaches is performed using publicly available and widely used databases.

REV Journal on Electronics and Communications, 2013

In computer vision many studies have been conducted in order to perform the matching and comparison of 3D models of objects. The main goal of matching is to group the models into different categories according to their similarity in order to allow their retrieval for recognition purposes and for further usage. So, in most of the cases, the comparison is run on a large dataset containing various models whether they belong to the same type of object or not and generally having similar or different shapes and poses. The objects’ nature and characteristics are important factors to be taken into consideration before performing the comparison step. We distinguish between two main categories of objects: rigid objects and deformable objects whose treatment and handling differ in the modeling as well as in the comparison phases. In this paper, we will be focusing on the comparison of deformable objects, and thus dealing with objects whose shapes might vary in different instances. For this pu...

2008

We present a novel 3D object retrieval method that relies upon a hybrid descriptor which is composed of 2D features based on depth buffers and 3D features based on spherical harmonics. To compensate for rotation, two alignment methods, namely CPCA and NPCA, are used while compactness is supported via scalar feature quantization to a set of values that is further compressed using Huffman coding. The superior performance of the proposed retrieval methodology is demonstrated through an extensive comparison against state-of-the-art methods on standard datasets.

IEEE Transactions on Multimedia, 2000

This paper proposes a novel framework for 3-D object retrieval, taking into account most of the factors that may affect the retrieval performance. Initially, a novel 3-D model alignment method is introduced, which achieves accurate rotation estimation through the combination of two intuitive criteria, plane reflection symmetry and rectilinearity. After the pose normalization stage, a low-level descriptor extraction procedure follows, using three different types of descriptors, which have been proven to be effective. Then, a novel combination procedure of the above descriptors takes place, which achieves higher retrieval performance than each descriptor does separately. The paper provides also an in-depth study of the factors that can further improve the 3-D object retrieval accuracy. These include selection of the appropriate dissimilarity metric, feature selection/dimensionality reduction on the initial low-level descriptors, as well as manifold learning for re-ranking of the search results. Experiments performed on two 3-D model benchmark datasets confirm our assumption that future research in 3-D object retrieval should focus more on the efficient combination of low-level descriptors as well as on the selection of the best features and matching metrics, than on the investigation of the optimal 3-D object descriptor.