Fattah Alizadeh - Transfer Talk - 22nd Oct 2012

Video Category: 
Transfer Talk
Fattah Alizadeh

Title: Design and Evaluation of a Partial Retrieval System for 3D Models Supporting Relevance Feedback

Abstract: As the number of 3D models is growing on the internet and other domain-specific datasets, the search and retrieval of such models are attracting a lot of attention in the scientific and even business domains. The basic and major part of any typical 3D content-based retrieval system is the shape descriptor. Accordingly, a great deal of research has been conducted to propose new descriptors and a lot of them have reasonable discriminative ability. But in addition to the retrieval quality, an ideal descriptor should tackle other existing problems caused by noise, simplification, arbitrary alignment of models, articulation and deformation. These main problems are generally ignored beyond achieving higher retrieval ability.

Furthermore, most of the available 3D retrieval systems perform matching based on the resemblance of the whole query to the available models. It is a challenging problem when, for example, the user only has some part of the desired model and looks for models with similar parts. Partial Matching is a possible solution to such a problem.

Finally, as the diversity of 3D models increases, automatically selected features may not be in accordance with the user's needs. Using Relevance Feedback (RF) users can overcome this problem and bridge the semantic gap between the abstract, high-level user intention and the low-level data representation and processing.

In the current research we have formulated the research questions by which we will try to solve all of the above mentioned problems. To date, we have solved the first research questions and proposed two different shape descriptors to obtain all of desired characteristics of an ideal shape descriptor. The experimental results of our descriptors show promising results compared to other state-of-the-art approaches.

To answer the other research questions and solve the problems of partial matching and relevance feedback, we will try to come up with new appropriate approaches by which the aforementioned problems are considered and solved properly.