摘要
研究了计算机视觉中扭曲目标识别问题,提出了获取仿射不变描述子的独立元分析方法,从而实现扭曲目标识别。该方法可直接对目标外形坐标数据进行处理,在仿射变换下具有不变性。通过仿真实验的方法给出了在不同扭曲度情况下,本方法和Fourier方法的识别数据。结果表明,在二维扭曲目标识别上,该方法比经典的Fourier方法具有更好的鲁棒性和更高的识别率。
The contorted object recognition is in computer vision. The affine invariant description has been obtained by uing ICA, and the contour target recognization was achived. The method has invariant characteristic under any affine transformation, and it coordinates of the contour of directly. It can be seen that the method has better performance in both robustness and recognition accuracy than Fouier methhod for 2D contorted objects recognition.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2004年第4期454-456,500,共4页
Systems Engineering and Electronics
基金
国家自然科学基金资助课题(60171036)
关键词
独立元分析
目标识别
仿射变换
independent component analysis
object recognition
affine transformation
