基于多尺度曲率植物叶片特征提取

基于B-样条小波计算边缘曲线多尺度曲率函数，根据多尺度信息筛选和定位超过一定曲率闲值的精角点，这样的点代表了边缘曲线的主要信息．文中使用了Canny边缘检测算子和数学形态学方法进行图像预处理，B-样条小波降低对噪声及扰动的灵敏性，以提高真实精角点定位水平．综合1，2]给出新的角特征矢量，并生成角点特征序列CS和弧段特征序列SS．特征序列可作为自适应-时滞单元混合神经网络的输入，通过学习完成图像分类与识别，对基于植物叶片形状识别种类提供辅助。

Based on Multiscale Curvat ure Feature Extracting of Plant Leaf

In this paper, based on B-spline wavelets multiscale curvature functions of image edges are calculated. By the multiscale curvature corners with important informations are filtered and located whose values are over a threshold. Some technologies, Canny edge detecting operating and Mathematical morphology are employed for image preprocessing and B-spline wavelets are employed for reducing the effect of noise or small variation, which brings us benefits to locate corner points with high accuracy . Basing on 1, 2] new corner feature vetor were given , and by which corner feature series CS and curve segment feature series SS are constructed. The numerical features can be as an input dataset of Adaptive-Delay Cell Hybrid Neural Network(ADCHNN), and image classification and recognition can be accomplished by learning. The work will give great help for plant identification by their leaf shape.