Abstract:

This paper presented an unsupervised texture segmentation algorithm utilizing EMD(Empirical Mode Decomposition) refined quaternionic spectrum. The innovations include: first, 2-D separable EMD algorithm was utilized to enhance the separability of features in frequency-domain; second, a quaternionic Gabor filter bank was designed to extract analytic features; at last, a novel validity function of FCM(Fuzzy C-Means Clustering) was introduced to realize unsupervised texture segmentation. Experiment results indicate that this method explores a reasonable way to use EMD and quaternionic spectrum analysis in image processing.

Key words: texture segmentation, quaternion, EMD(Empirical Mode Decomposition), Gabor filter, time-frequency analysis

摘要：

提出一种基于EMD细化四元数谱的无监督纹理分割方法。创新之处在于:首先采用二维可分离EMD分解提高各图像成分的四元数Fourier频域特征的分离度;再使用四元数Gabor滤波器组进行解析特征提取;最后利用新的模糊C均值聚类评价函数实现无监督的纹理分割。实验结果证明了该方法的有效性。

关键词: 纹理分割, 四元数, EMD分解, Gabor滤波器, 时频分析