Fig 1 - uploaded by Silvia Biasotti
Content may be subject to copyright.
Source publication
We describe a 3D shape classification framework, and discuss the performance of selective and creative prototypes extracted
from structural descriptors.
Context in source publication
Context 1
... methods establish the membership of an unknown query shape in one of a set of classes, thus inferring semantic information about the query model. In this paper, we discuss the role of 3D prototypes for shape classi- fication. Prototypes are embedded in a general, dissimilarity-based classification framework. The flow is illustrated in Fig. 1. For each class, a small set of proto- types is defined (a); a query is classified at run-time by matching its descriptor vs. the subset of prototypes (b), thus reducing the search space. Prototypes can be defined in either a selective or a creative manner. In the first case, one or a few class members are chosen to represent the whole ...
Similar publications
This paper addresses two issues hindering the advances in accurate image alignment. First, the performance of descriptor-based approaches to image alignment relies on the chosen descriptor, but the optimal descriptor typically varies from image to image, or even pixel to pixel. Second , the neighborhood structure for smoothness enforcement is usual...
Architectures based on composing target application functionality from pre-existing components have been successfully used in many projects, yet there are several aspects in which they fail to reach the desirable level of maturity. Since different vendors may provide components with the same functionality, extra-functional properties must be taken...
The segmentation of images is a critical step in many computer vision applications. Besides, some applications require to achieve acceptable segmentation quality, whilst the algorithm is executed in real time. In this study, we present a split-and-merge segmentation methodology that uses integral images to improve its execution time. We called our...
State-of-the-art texture descriptors typically operate on grey scale images while ignoring color information. A common way to obtain a joint color-texture representation is to combine the two visual cues at the pixel level. However, such an approach provides sub-optimal results for texture categorisation task.
In this paper we investigate how to op...
Citations
... A survey of three typical semantics processing techniques (relevance feedback, machine learning, and ontology) is presented in [71]. Typical semantics-based 3D retrieval approaches include relevance feedback [72], semantic labeling [73], neural networks [74], supervised [75][76][77][78] or semi-supervised [79][80][81] learning, boosting [82], prototypes [83], autotagging [84], spectral clustering [85], manifold ranking [86], semantic tree [87], feature dimension reduction [88], semantic subspaces [89], class distances [54], semantics annotation of 3D models [90], semantic correspondences [91], and sparse structure regularized ranking [92]. ...
... Due to the variations in the types and robustness considerations in retrieval performance, we employ the hybrid shape descriptor ZFDR devised in [54] which integrates both visual and geometric information of a 3D model: Zernike moments and Fourier descriptor features of 13 cube-based sample views; Depth information feature of 6 depth buffer views and Ray-based features based on ray shooting from the center of the model to its farthest surface intersection points. Visual information-based features (e.g., Z and F) have good performance in characterizing some classes like ''sea animal'', but for some other types of models like ''car'', depth buffer-based features (e.g., D and R) are better [83]. We optimally integrate the above four different but complementary features to formulate the hybrid shape descriptor ZFDR to increase its differentiation power. ...
Large-scale 3D shape retrieval has become an important research direction in content-based 3D shape retrieval. To promote this research area, two Shape Retrieval Contest (SHREC) tracks on large scale comprehensive and sketch-based 3D model retrieval have been organized by us in 2014. Both tracks were based on a unified large-scale benchmark that supports multimodal queries (3D models and sketches). This benchmark contains 13680 sketches and 8987 3D models, divided into 171 distinct classes. It was compiled to be a superset of existing benchmarks and presents a new challenge to retrieval methods as it comprises generic models as well as domain-specific model types. Twelve and six distinct 3D shape retrieval methods have competed with each other in these two contests, respectively. To measure and compare the performance of the participating and other promising Query-by-Model or Query-by-Sketch 3D shape retrieval methods and to solicit state-of-the-art approaches, we perform a more comprehensive comparison of twenty-six (eighteen originally participating algorithms and eight additional state-of-the-art or new) retrieval methods by evaluating them on the common benchmark. The benchmark, results, and evaluation tools are publicly available at our websites (http://www.itl.nist.gov/iad/vug/sharp/contest/2014/Generic3D/, 2014, http://www.itl.nist.gov/iad/vug/sharp/contest/2014/SBR/, 2014).
... Xu and Li [46] defined the distance between two models by adding the weighted difference between their back propagation neural network (BPNN) 3D model classification output vectors and the Euclidian (L2) distance between their 3D moment feature vectors. Biasotti et al. [6] proposed a 3D model classification approach by comparing a query model with several prototypes selected to represent a class and applied the prototype-based scheme into a 3D model retrieval application. They made a comparison study of the Nearest Neighbor (NN)-based and two prototype-based retrieval methods and the results show that NN achieves the best retrieval accuracy, though may be slower. ...
To improve the retrieval performance on a classified 3D model database, we propose a 3D model retrieval algorithm based on
a hybrid 3D shape descriptor ZFDR and a class-based retrieval approach CBR utilizing the existing class information of the
database. The hybrid 3D shape descriptor ZFDR comprises four features, depicting a 3D model from different aspects and it
itself is already comparable to or better than several related shape descriptors. To compute the distance between a query
model and a target model within a class of a database, we define an integrated distance metric which takes into account the
class information. It scales the distance between the query model and the target model according to the distance between the
query model and the class. Our class-based retrieval approach CBR is general, it can be used with any shape descriptors to
improve their retrieval performance. Extensive generic and partial 3D model retrieval experiments on seven standard databases
demonstrate that after we employ CBR, the retrieval performance of our algorithm CBR-ZFDR is evidently improved and the result
is better than that achieved by the state-of-the-art method on each database in terms of most of the commonly used performance
metrics.
Keywords3D model retrieval–Hybrid shape descriptor–Class information–Integrated distance
This paper presents a fuzzy KNN and Bayesian decision based classification method for determining whether a 3D object belongs to human class. To achieve efficiency and simplicity, the view having maximum area is used to substitute a 3D shape, which is front-side view for human models. View features and structural features are utilized to describe those selected views. For shape feature can not distinguish one class from the others crisply, fuzzy KNN is adopted to estimate the probability that the object belongs to human class using view feature extracted from an unknown object and view features of training objects. Structural features are represented as adjacent graphs after shape decompositions, while characteristic structures are predefined for human class. Next priori probabilities that the characteristic structure exist in human and other classes are gained from training objects. For an unknown object, an adjacent graph is checked whether the characteristic structure exists, and then fuzzy classification results and priori probabilities are used together to classify the unknown object by Bayesian rule. Experimental results show that our approach have good accuracy for classifications.
