基于小波包框架及主成份分析的纹理图像分割

由于纹理图像的信息主要集中在中频区域,利用小波包框架分解纹理图像得到的子图既与原图的大小相同,同时也保留了中频信息,因此可以从子图中提取特征用于分割。为减少特征维数,在同一子通道中只保留具有最大方差的子图作为初始特征图,再从每一个初始特征图中分别提取多维特征。为去除特征中噪声的同时保持特征中的边缘,提出四分均值滤波器对特征滤波。为进一步减少特征间的冗余信息,降低聚类分析的计算负荷,采用主成份分析法不但保留了原有特征中的主要信息,而且使得特征维数有大幅度的减少。模糊c-均值聚类法实现了对于特征的聚类。利用该文的方法分割Brodatz纹理集中的纹理图组成的测试图,达到了较高的准确率。

Texture Segmentation Based on the Wavelet Packets Frame and Princiole Comoonent Analysis

The main information of texture images focuses in the middle frequency region and the wavelet packets frame can decompose the texture image into subimages that have the same size with the original image and contain the middle frequency information.In order to decrease the number of feature,only the one having the maximum variance in a subchannel is selected out to be the initial feature among all of the subimages.Then multiple features are extracted from every initial feature.We proposes to use the quadrant-mean filter to smooth the noise in the feature while alleviat-ing the boundary effect of the features.At last PCA method,which can decrease features greatly in dimension while keeping the main information only in few new features,is apply to decrease the burden of the clustering.The feature clustering is obtained by fuzzy c-means method.The performance of this new method is demonstrated on the Brodatz texture set and the segmentation correct ratio is up to95%.