Abstract: This paper proposes an algorithm based on the similarity measurement of the spectral histograms for texture segmentation.In the texture segmentation algorithm，the seed selection is not needed.First of all，to acquire the texture features，using the algorithm of filter selection，then based on a local spatial/frequency representation，spectral histogram is employed as feature statistic for texture segmentation.For another，the initial result is derived by using x²-statistic to measure the similarity between two spectral histograms.The final segmentation result is obtained by applying the method of morphology skeletonization to locate the region boundaries.Experimental results demonstrate that the proposed method is effective.

Key words: spectral histograms, feature extraction, x²-statistic, skeletonization, morphology, texture segmentation

摘要： 提出基于谱直图相似性的纹理图像分割算法，在提出的纹理分割算法中，无需选择种子点。首先通过使用一组滤波器来获取纹理图像的纹理特征，基于空域/频域的表示方法，使用谱直方图作为特征统计，然后利用x²统计特征度量谱直方图之间的相似性，得到初始分割，然后基于形态学骨架化的原理，对区域边界进行精确定位，得到最终的分割结果。实验结果验证了算法的有效性。

关键词: 谱直方图, 特征提取, x²统计特征, 骨架化, 形态学, 纹理分割