Abstract: This paper presents an unsupervised texture segmentation algorithm in quaternion-domain based on empirical mode decomposition. This method uses 2-D separable EMD to enhance the separability of features in frequency-domain, then a quaternionic Gabor filter bank is designed to extract analytic features and a novel validity function of fuzzy c-means clustering is introduced to realize texture segmentation. Experimental results indicate the effectiveness of the method.

Key words: Texture segmentation; Quaternion; Empirical mode decomposition(EMD); Gabor filter; Time-frequency analysis

摘要： 提出了一种基于EMD 分解的四元数域无监督纹理分割方法。首先采用二维可分离EMD 分解提高各图像成份的四元数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。实验结果验证了该方法的有效性。

关键词: 纹理分割；四元数；经验模式分解；Gabor 滤波器；时频分析