基于FTMP特征的小波域非监督纹理分割新算法

A Novel Unsupervised Texture Segmentation Algorithm Based on FTMP in Wavelet Domain

为了能够在非监督环境下利用有限纹理混合模式(finite texture mixture pattern,FTMP)特征进行图像分割,提出了一种基于两步聚类的小波域非监督纹理分割算法。该算法通过两步Kmeans聚类来实现多尺度纹理分割。第一步Kmeans聚类计算每一子带的PLVP及对应的FTMP特征;第二步Kmeans则对每一尺度的FTMP特征进行聚类,从而计算各尺度的分割结果。为了获取更为可靠的分割结果,算法考虑了不同尺度之间的交互。合成纹理影像和遥感影像的分割实验验证了该算法的有效性。

In order to segment textures in the unsupervised circumstance based on the finite texture mixture pattern （ FTMP）, a new unsupervised texture image segmentation algorithm was proposed based on the two - step clustering in wavelet domain. It employed a two - step Kmeans algorithm to complete the mutiscale texture segmentation. The first - step Kmeans calculated the PLVP and the FTMP feature on each scale. And the second - step Kmeans obtained the segmentation result on each scale by the clustering of FTMP feature. In order to achieve more reliable segmentation results,interactions between different scales were considered. The effectiveness of this method was testified by experiments on both synthetic texture images and remote sensed images.