基于双树复数小波变换的微钙化诊断方法

提出一种基于双树复数小波变换的微钙化分类方法.通过提取基于小波和灰度直方图的纹理特征,结合遗传算法进行特征优化,分别用神经网络,支持向量机和KNN分类器进行微钙化的良恶性分类.对三种不同的分类器进行对比,结果表明:KNN分类器取得最好的效果,而支持向量机优于神经网络.KNN分类器对比于神经网络和支持向量机,无需训练,可节约训练时间,最直接地利用了样本和样本之间的关系,减少了类别特征选择不当对分类结果造成的不利影响,可以最大程度地减少分类过程中的误差项.在类别决策时,KNN分类器只与极少量的相邻样本有关,可以较好地避免样本数量的不平衡问题.与传统的小波比较,双树复数小波具有近似平移不变性和正则性,对图像信号具有良好的方向选择性,且冗余度有限,计算量较小.

Microcalcification Diagnosis Based on Dual-Tree Complex Wavelet Transform

A new diagnosis method of microcalcification is applied based on dual-tree complex wavelet transform.A set of texture features are extracted,including the wavelet based features and gray-level histogram based features.Combining the genetic algorithm to optimize the features,the neural network,support vector machine and KNN classifier are used to distinguish the benign and malignant microcalcifications.The results of the three different classifiers are analyzed,which show that KNN classifier get the best result and support vector outperformed neural network.Compared with the other two classifiers,KNN classifier is a non-training method which can save much training time.It directly computes the distances between samples,which can reduce the adverse effects caused by inefficient selection of features and minimize the error in classification.The KNN classifier depends on a small quantity of samples when kind deciding,which can solve the problem caused by the imbalance of samples.Compared to traditional wavelet basis,the shift invariance and high regularity of dual-tree complex wavelet make it more helpful for direction selection of image signal,and dual-tree complex wavelet transform will get limited redundancy and less amount of calculation.