基于QWT和GLCM的多特征双重加权纹理分割

摘要

针对传统特征加权算法对混合属性数据只进行全局样本加权，而忽略不同特征提取算法对纹理描述性能强弱的缺点，提出一种基于算法有效性和特征重要性的双重加权策略。将四元数小波变换和灰度共生矩阵融合特征应用k-means算法进行初聚类，并以此产生的初始聚类中心作为参考，取每类聚类中心的k-近邻样本作为双重加权的训练样本集合。利用改进的ReliefF算法和相关性度量解决特征内权值的设定问题，再利用SVM解决特征间加权问题，最后将双重特征加权结合FCM应用于纹理分割。实验结果表明，该方法在合成纹理和自然纹理图像中均有较好的性能，且较其他特征加权算法分割准确度更高。%Traditional feature weighting algorithms only weight sample globally for mixed attribute data, which ignores the fact that different feature extraction methods are suited to extract different aspects of texture feature. Therefore, a dual weighted strategy is proposed based on the validities and feature importance. Firstly, the fused features of quaternion wavelet transform and gray level cooccurrence matrix are clustered by the k-means algorithm, and the initial cluster centers are regarded as a reference. The k-nearest neighbor samples extracted from each cluster center are regarded as double weighted training sample sets. Then, the problem of weights inside feature is solved by using modified ReliefF algorithm and correlation measure, and the problem of weights between features is solved by using Support Vector Machine. The experimental results show that the proposed method has a good performance in synthetic textures and natural texture images, and has higher segmentation accuracy than other feature weighting algorithms.