多尺度目标识别新方法

利用全局矩反映的是图像在整个空间的统计特征,可以稳定可靠地识别形状差异较大物体的特点,以及小波矩描述的是图像在局部空间的特征,能够准确有效地区分形状相似的目标的优点,提出了一种新的由粗到细、从全局到局部的多尺度逐级匹配目标识别方法.这种在大尺度下采用全局特征进行粗分类,小尺度下采用局部特征进行细分类的思想加快了匹配过程,实验结果表明,该方法比仅使用全局不变矩或小波不变矩特征的目标识别方法具有更高的识别率和更广的适用范围.

A New Multiresolution Target Recognition Method

General moments and wavelet moments are utilized in the recognition method presented in this paper, because the former can recognize the patterns with obvious differences for their statistic character and the latter can distinguish local nuances from similar targets using their multiresolution analysis. The method matches patterns using multiresolution analysis from coarse to exact, from whole to local. This kind of thoughts expedites the matching process that differ the whole in big resolutions coarsely and the local in small ones exactly. The experimental results indicate that with this method the recognition property is much higher than that of only using general moments or wavelet moments.