EG 3DOR 08 Program
|9:15 - 9:30
Session 1 - 3D SHAPE DESCRIPTORS
9:30 - 10:30
- Characterizing Shape Using Conformal Factors
Mirela Ben-Chen and Craig Gotsman
We present a new 3D shape descriptor based on conformal geometry. Our descriptor is invariant under non-rigid
quasi-isometric transformations, such as pose changes of articulated models, and is both compact and efficient to
compute. We demonstrate the performance of our descriptor on a database of watertight models, and show it is
comparable with state-of-the-art descriptors.
- 3D Object Retrieval using an Efficient and Compact Hybrid Shape Descriptor
Panagiotis Papadakis, Ioannis Pratikakis, Theoharis Theoharis, Georgios Passalis and Stavros Perantonis
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.
10:30 - 11:00
Session 2 - RETRIEVAL BY PARTS
11:00 - 12:30
- Isometry-invariant matching of point set surfaces
Mauro R. Ruggeri and Dietmar Saupe
Shape deformations preserving the intrinsic properties of a surface are called isometries. An isometry deforms a
surface without tearing or stretching it, and preserves geodesic distances. We present a technique for matching
point set surfaces, which is invariant with respect to isometries. A set of reference points, evenly distributed on
the point set surface, is sampled by farthest point sampling. The geodesic distance between reference points is
normalized and stored in a geodesic distance matrix. Each row of the matrix yields a histogram of its elements.
The set of histograms of the rows of a distance matrix is taken as a descriptor of the shape of the surface. The
dissimilarity between two point set surfaces is computed by matching the corresponding sets of histograms with
bipartite graph matching. This is an effective method for classifying and recognizing objects deformed with
isometric transformations, e.g., non-rigid and articulated objects in different postures.
- Markov Random Fields for Improving 3D Mesh Analysis and Segmentation
Guillaume Lavoue and Christian Wolf
Mesh analysis and clustering have became important issues in order to improve the efficiency of common processing
operations like compression, watermarking or simplification. In this context we present a new method for
clustering / labeling a 3D mesh given any field of scalar values associated with its vertices (curvature, density,
roughness etc.). Our algorithm is based on Markov Random Fields, graphical probabilistic models. This Bayesian
framework allows (1) to integrate both the attributes and the geometry in the clustering, and (2) to obtain an optimal
global solution using only local interactions, due to the Markov property of the random field. We have defined
new observation and prior models for 3D meshes, adapted from image processing which achieve very good results
in terms of spatial coherency of the labeling. All model parameters are estimated, resulting in a fully automatic
process (the only required parameter is the number of clusters) which works in reasonable time (several seconds).
- Part Analogies in Sets of Objects
Shy Shalom, Lior Shapira, Ariel Shamir and Daniel Cohen-Or
Shape retrieval can benefit from analogies among similar shapes and parts of different objects. By partitioning an
object to meaningful parts and finding analogous parts in other objects, sub-parts and partial match queries can
be utilized. First by searching for similar parts in the context of their shape, and second by finding similarities even
among objects that differ in their general shape and topology. Moreover, analogies can create the basis for semantic
text-based searches: for instance, in this paper we demonstrate a simple annotation tool that carries tags of
object parts from one model to many others using analogies. We partition 3D objects based on the shape-diameter
function (SDF), and use it to find corresponding parts in other objects. We present results on finding analogies
among numerous objects from shape repositories, and demonstrate sub-part queries using an implementation of a
simple search and retrieval application.
13:00 - 14:30
Session 3 - LEARNING
14:30 - 15:30
- Similarity Score Fusion by Ranking Risk Minimization for 3D Object Retrieval
Ceyhun Burak Akgul, Bulent Sankur, Yucel Yemez, and Francis Schmitt
In this work, we introduce a score fusion scheme to improve the 3D object retrieval performance. The state of
the art in 3D object retrieval shows that no single descriptor is capable of providing fine grain discrimination required
by prospective 3D search engines. The proposed fusion algorithm linearly combines similarity information
originating from multiple shape descriptors and learns their optimal combination of weights by minimizing the
empirical ranking risk criterion. The algorithm is based on the statistical ranking framework [CLV07], for which
consistency and fast rate of convergence of empirical ranking risk minimizers have been established. We report the
results of ontology-driven and relevance feedback searches on a large 3D object database, the Princeton Shape
Benchmark. Experiments show that, under query formulations with user intervention, the proposed score fusion
scheme boosts the performance of the 3D retrieval machine significantly.
- A Neurofuzzy Approach to Active Learning based Annotation Propagation for 3D Object Databases
Michalis Lazaridis and Petros Daras
Most existing Content-based Information Retrieval (CBIR) systems using semantic annotation, either annotate all
the objects in a database (full annotation) or a manually selected subset (partial annotation) in order to increase
the system's performance. As databases become larger, the manual effort needed for full annotation becomes un-
affordable. In this paper, a fully automatic framework for partial annotation and annotation propagation, applied
to 3D content, is presented. A part of the available 3D objects is automatically selected for manually annotation,
based on their "information content". For the non-annotated objects, the annotation is automatically propagated
using a neurofuzzy model, which is trained during the manual annotation process and takes into account the infor-
mation hidden into the already annotated objects. Experimental results show that the proposed method is effective,
fast and robust to outliers. The framework can be seen as another step towards bridging the semantic gap between
low-level geometric characteristics (content) and intuitive semantics (context).
15:30 - 16:00
Session 4 - APPLICATIONS - 3D FACE RECOGNITION
16:00 - 17:00
- Face Recognition by SVMs Classification and Manifold Learning of 2D and 3D Radial Geodesic Distances
Stefano Berretti, Alberto Del Bimbo, Pietro Pala and Francisco Jose Silva Mata
An original face recognition approach based on 2D and 3D Radial Geodesic Distances (RGDs), respectively
computed on 2D face images and 3D face models, is proposed in this work. In 3D, the RGD of a generic point
of a 3D face surface is computed as the length of the particular geodesic that connects the point with a reference
point along a radial direction. In 2D, the RGD of a face image pixel with respect to a reference pixel accounts for
the difference of gray level intensities of the two pixels and the Euclidean distance between them. Support Vector
Machines (SVMs) are used to perform face recognition using 2D- and 3D-RGDs. Due to the high dimensionality of
face representations based on RGDs, embedding into lower-dimensional spaces using manifold learning is applied
before SVMs classification. Experimental results are reported for 3D-3D and 2D-3D face recognition using the
- A 3D Face Recognition Algorithm Using Histogram-based Features
Xuebing Zhou, Helmut Seibert, Christoph Busch and Wolfgang Funk
We present an automatic face recognition approach, which relies on the analysis of the three-dimensional facial
surface. The proposed approach consists of two basic steps, namely a precise fully automatic normalization stage
followed by a histogram-based feature extraction algorithm. During normalization the tip and the root of the nose
are detected and the symmetry axis of the face is determined using a PCA analysis and curvature calculations.
Subsequently, the face is realigned in a coordinate system derived from the nose tip and the symmetry axis, resulting
in a normalized 3D model. The actual region of the face to be analyzed is determined using a simple statistical
method. This area is split into disjoint horizontal subareas and the distribution of depth values in each subarea is
exploited to characterize the face surface of an individual. Our analysis of the depth value distribution is based
on a straightforward histogram analysis of each subarea. When comparing the feature vectors resulting from the
histogram analysis we apply three different similarity metrics. The proposed algorithm has been tested with the
FRGC v2 database, which consists of 4950 range images. Our results indicate that the city block metric provides
the best classification results with our feature vectors. The recognition system achieved an equal error rate of
5.89% with correctly normalized face models.
Session 5 - PLATFORMS
17:00 - 17:30
- On-line and open platform for 3D object retrieval
Benoit Le Bonhomme, B. Mustafa, Sasko Celakovsky, Marius Preda, Francoise Preteux and D. Davcev
In this paper we present the MyMultimediaWorld Internet-based platform designed to benchmark descriptors and
description scheme for 3D object retrieval purpose. Relying on the MPEG-4 and MPEG-7 multimedia standards
for data representation and description respectively, this open platform is designed to host multiple datasets, descriptors,
descriptor extraction algorithms and similarity measures. We implemented an easy-to-use API designed
to make the integration of the 3D object retrieval technology of third-party researchers agnostic to and independent
of the global system complexity. Benchmarking results are automatically updated accordingly and presented
qualitatively by displaying the 3D retrieved objects and quantitatively by providing the estimates of the state-of-the
art performance criteria.
Round Table Discussion "Future trends in 3D Object Retrieval"
SHORT PAPERS (POSTER PRESENTATION)
12:30 - 13:00
- A Novel Approach for Range Image to 3D Model Partial Matching
Georgios Stavropoulos, Konstantinos Moustakas, Dimitrios Tzovaras and Michael G. Strintzis
This paper presents a novel method for 3D model similarity search based on a query-by-range-image approach.
A technique for matching the salient features between the 3D model and the range image is proposed, followed
by fast and reliable hierarchical similarity search in the parameter space. Experimental results are provided on a
benchmarking database that show the very good performance of the method.
- Global 3D Mesh Segmentation Using Local Operators
Indriyati Atmosukarto and Linda G. Shapiro
Mesh segmentation is an important operation both in computer vision and computer graphics. Regions obtained
from mesh segmentation are used in various applications such as 3D object retrieval, 3D object recognition,
and 3D mesh editing. We present a new mesh segmentation technique that uses local operators to obtain global
segmentation of the 3D mesh. The technique is based on a four-phase pipeline that provides the flexibility to
incorporate different low-level features and mid-level feature aggregation methods. Experimental results show
that our method produces patches that show repeatability across objects of the same class and will therefore be
useful for object recognition tasks.
- A 3D Pottery Database for Benchmarking Content Based Retrieval Mechanisms
Anestis Koutsoudis, George Pavlidis, Fotios Arnaoutoglou, Despoina Tsiafakis and Christodoulos Chamzas
The benchmarking of 3D content base retrieval mechanisms is usually performed on test bed datasets composed
by 3D models. These usually cover several categories of objects such as ships, airplanes, animals, furniture, etc.
In this paper, we attempted the generation of a ground truth database of 3D models that exhibit morphological
characteristics similar to those found in ancient Greek pottery. We developed a software tool to model 3D vessels
based on bitmap profile images, enhanced with the function to semi-automatically generate random 3D vessels
accompanied by metadata. The metadata follow a proposed MPEG-7 compatible schema which covers the basic
information required by an archaeologist to describe a vessel and the MPEG-7 3D Shape Spectrum Descriptor
for allowing possible performance comparisons of novel descriptors against a standard.
- 3D Object Retrieval based on Resulting Fields
Athanasios Mademlis, Petros Daras, Dimitrios Tzovaras and Michael G. Strintzis
3D object search and retrieval has become a very challenging research field over the last years with application
in many areas like computer vision, car industry, medicine, etc. All approaches that have been proposed so far are
based on the analysis of the shape of the 3D object, either concerning its surface or its volume. In this paper, a
completely different approach is followed: Instead of extracting features from the 3D object, its shape's "impact"
on the surrounding area is examined. This impact is expressed by considering the 3D object voxels as electric
point charges and computing the resulting electrostatic field in a neighborhood around it. The proposed approach
ensures robustness with respect to object's degeneracies and native invariance under rotation and translation.
Experiments which were performed in a 3D object databases proved that the proposed method can be efficiently
used for 3D object retrieval applications.