扭对称检测的新进展

RECENT PROGRESSES IN SKEWED SYMMETRY DETECTION

对称是目标的基本形状特征,基于对称的形状表示和目标识别技术是模式识别和人工智能领域的重要研究方向。3-D空间中目标的对称表面在图像平面的投影往往呈现扭对称形态,利用扭对称信息可以加快诸如目标的姿态估计、方向计算以及图像校准等过程。本文详细地介绍了扭对称的形成模型、表示技术以及近来出现的三类主要的检测方法,即基于Hough变换的方法、基于不变特征的方法和基于规格化的方法,我们在叙述它们的主要思想、所适用的数据类型和优缺点后,给出了进一步研究的可能方向。

Symmetry is a basic shape feature of objects. Shape representation and object recognition based on symmetry is an important research area in patttern recognition and artificial intelligence. The projections of the symmetrical surfaces of 3-D objects always show as skewed-symmetry in image plane. Using skewed symmetry information we can quicken the procedures such as object pose estimation, direction computation and image registration. In this paper, we describe in detail the forming models and representation techniques of skewed symmetry and then introduce three newly appeared main detection approaches, which are the one based on Hough transform, the one based on invariant signatures and the one based on regularization. We also give the possible directions of further studies after we introduce their main ideas, the data to be processed, and their merits and shortcomings.