摘要
本文在充分分析了现有各种仿射不变量的基础上 ,提出了一种新的统计仿射不变量 ,它对于发生尺度变化、旋转、扭曲和平移的目标具有不变性。实验结果表明 ,相对于传统的仿射不变量 ,在目标轮廓分割不完整或噪声污染的情况下 ,仍能够保持较高的稳健性 ,同时克服了基于轮廓的小波方法和傅立叶级数等方法对目标轮廓出现缺陷时的不足 ,在图像目标识别中具有实用价值。
In this paper a new statistical affine invariant, which has invariance for the same image object under scale, rotation, distortion and translation changes, is proposed based on the research of traditional affine invariants. Test results show that the statistical affine invariant can keep highly stable even if the object contour is segmented badly or is noisy as compared with other invariants, thus can conquer the deficiency that exists in the wavelet affine invariant and the Fourier series invariant when the ill-segmented contour occurs. In real object recognition, the proposed statistical affine invariant will have a broad prospect.
出处
《计算机工程与科学》
CSCD
2004年第9期35-38,47,共5页
Computer Engineering & Science
