用不变矩实现花粉粒显微图像的归一化

Image normalization for the pollen granule of herbal flowers with invariant moments

显微图像易受光照、视角、方位、噪声等的影响。在这些因素的作用下,同一类显微图像的不同变形体差距有时大于另一类图像,因此进行显微图像识别以前需要进行图像的归一化。本文推导了基于变形雅可比(p=4,q=3)-傅立叶矩的草药花粉粒显微图像的归一化标准,并对8种中蒙花药显微特征图像进行归一化实验,发现同一种花粉粒图像不同变形体归一化后的矩值方差明显小于归一化前。另外,用加权最小平均距离分类器对320个花粉粒归一化显微图像进行初步识别实验,平均识别率达97.4%。

The image normalization criteria of Herbal pollen granule have been derived by using Pseudo-Jacobi-Fourier Moments （PJFM＇s） in this paper. The normalizing experiments have been carried out for pollen granule microscopic characteristics of 8 kinds of Herbal flowers,and 320 normalized pollen granule microscopic characteristics have been classified by using the weighted minimum-mean-distance rule. The results show that the standard deviation of invariant moments of normalized microscopic images is significantly smaller than that of non-normalized images,and the average classification rate reached as high as 97.4%.