11-12-2009, 09:44 PM
In this paper a novel method for 3D content-based search and retrieval is proposed. Weighted 3D Krawtchouk moments are introduced for efficient 3D analysis which are suitable for content-based search and retrieval applications.
Introduction
3D search finds application in gaming, modelling, manufacturing and molecular biology applications.
Previous proposals for 3D search:
-a fast query by example approach where the descriptors are properly chosen in order to follow the basic geometric criteria which humans usually use for the same purpose.
-a method based on shape histograms.
-based on Generalized Radon Transform(GRT).
-a web-based 3D search and retrieval system
-a visual similarity-based 3D model retrieval system.
The drawback of these methods is the loss of discriminative power.
Weighted 3D Krawtchouk
Given a 3D object as input, the Weighted 3D Krawtchouk moments are computed, which are then used as a descriptor vector. Thus a very compact description vector is made. The descriptor extraction is very fast and the matching process, one-to-all, for a single object in a medium size database can be completed in few seconds.The method is not invariant under geometrical transformation, thus for every query 3D model a preprocessingpose and position normalization step is required.
Seminar report download:
[attachment=585]
Introduction
3D search finds application in gaming, modelling, manufacturing and molecular biology applications.
Previous proposals for 3D search:
-a fast query by example approach where the descriptors are properly chosen in order to follow the basic geometric criteria which humans usually use for the same purpose.
-a method based on shape histograms.
-based on Generalized Radon Transform(GRT).
-a web-based 3D search and retrieval system
-a visual similarity-based 3D model retrieval system.
The drawback of these methods is the loss of discriminative power.
Weighted 3D Krawtchouk
Given a 3D object as input, the Weighted 3D Krawtchouk moments are computed, which are then used as a descriptor vector. Thus a very compact description vector is made. The descriptor extraction is very fast and the matching process, one-to-all, for a single object in a medium size database can be completed in few seconds.The method is not invariant under geometrical transformation, thus for every query 3D model a preprocessingpose and position normalization step is required.
Seminar report download:
[attachment=585]
