Optimization of feature representation for point cloud 3D models based on evolutionary computation

Abstract

Three dimensional objects are an important type of multimedia data with many promising applications: criminal identifications, computer animation, movie industry and medical industry. Therefore, over the past ten years, with the fast development of scanning technologies and other modeling techniques, 3D models grow dramatically in number and receive increasing attention for a number of reasons: advancement in 3D hardware and software technologies, the decreasing prices and increasing availability of the models, more affordable 3D authoring tools, and the formulation of open standards for 3D data interchange. Against this background, a number of approaches for 3D model classification and retrieval have been proposed and several 3D model classification and retrieval systems for Internet or CAD applications have also been developed. However, most of the previously developed techniques for 3D model classification are based on the original polygonal representation. In particular, the feature representation of each model is based on surface properties such as the neighborhood relationship between vertices, and the local surface orientation, which are difficult to estimate in the case of the models based on point cloud, an expressive and intuitive approach to describe a 3D model. As a result, we need to develop new techniques for solving the problem. In view of this requirement, we propose a new feature representation for 3D point cloud models based on a set of principal projection axes. The point set is then projected onto each of these axes, and a suitable summary statistics of the projected point set along each axis is calculated. The complete set of statistics is then adopted as the feature representation of the point set. We propose to adopt Evolutionary Strategy (ES) as the optimization technique. This is in view of the capability of ES to explore most regions of the search space in parallel through the generation of a large number of potential candidate solutions in a population. The effectiveness of each solution is measured through a fitness function which represents our current optimization criterion, and individuals associated with high fitness values are allowed to reproduce through the operations of recombination and mutation, while those with low fitness values are displaced out of the population. In this way, highly effective solutions will gradually emerge through this process of competition and selection. In our approach, we transform this problem from 3D to 1D, which is easier to solve. Subsequently, we extend this method by transforming it from 3D to 2D, which is a plane. In this way, we can take advantage of all the techniques of image processing to improve the effectiveness further. To evaluate our approaches, we focus on 3D head model classification problem.

Date of Award	16 Jul 2007
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Hau San WONG (Supervisor)