改进的主动形状模型在植物叶形分类上的应用

本文在继承主动形状模型以地标点序列表示对象形状轮廓，通过归一化序列间的统计关系来研究形状间的相似性和差异性，以及用识别率来体现识别效果这三个基本特征的同时，针对植物叶片识别的特点，对模型进行了改进：增加了独立参数和粗分类步骤，使叶片识别免受非形状分类因素的干扰；在归一化进程中分别应用覆盖面积半径和生物学参照点进行伸缩变换和旋转变换，使归一化更符合生物学特征；在分类时通过设置相似度阈值，强化了同类叶片的标准，从而减少了一些非同类的误归入事件的发生；最后设计了三组实验来研究各影响因素独立变化时对识别率的影响，同时引入了误差矩阵和相似度矩阵，分别来体现错分的事件的分布情况并揭示了产生错分的原因。经过这些改进，本文实现植物叶片图像的批量识别和分类，提供了一种查询错分分布的方法并指出提高识别率的研究方向。

Advanced Active Shape Model Applications on Classifying Plant Leaves

The purpose of this paper is to recognize and classify plant-leaf images automatically in batches. Improvements on Active Shape Model are made which include adding independent coefficient and a rough classifying stage before the images are processed, taking advantage of area-covering radius and peak reference point at the aligning stage, and setting up a threshold value of similarity during classifying. Three groups of experiments are designed to later measure the influence from each affecting variable as well as to test the rate of accurately recognizing as a whole, in which error-matrix and similarity-matrix are raised up to describe the distribution of error events and to appeal the reason for errors. Such data points to a further improvement for ASM models.