基于图像纹理和小波变换的聚类分割方法  

A Method of Image Clustering Segmentation Based on Image Texture and Wavelet Transform

作　　者：张勇[1] 周文卫[1] 陈立杨[1] 石高峰[1] 

机构地区：[1]西南交通大学电气工程学院,四川省成都市610031

出　　处：《自动化信息》2010年第7期48-49,63,共3页Automation Information

摘　　要：针对纹理图像，本文提出了一种基于图像纹理特征的非学习分割方法。采用小波变换和快速k-means聚类分割算法，减少了整个处理过程的运算量。为了保证分类算法的精确性，运用了总体流量变化最小（Total Variation Flow）[1]非线性去噪方法对图像进行预处理，从而将减小图像噪声污染带来的分割误差。在图像特征的提取上，运用Gabor滤波器原理生成滤波空间，并让图像通过滤波空间而生成特征向量空间。通过制定一个快速寻优策略，从而达到分割图像的目的。Aiming at textured images, the authors propose a non-learning image segmentation method based on image texture fea- tures in this paper. The computation workload of the whole process is reduced by means of wavelet transform and fast k-means clustering segmentation algorithm. In order to ensure the accuracy of classification algorithm, a non-linear image-denoising method of ＂ Total Variation Flow minimality＂ is adopted to implement the image-preprocessing, thus the segmentation error which is caused by image noise pollution will be decreased. In the image feature extracting process, a filtering space is generated by using Gabor filters principle and the test images will generate the feature vector space through the filtering space. A fast optimization searching strategy is formulated to achieve the purpose of image segmentation.

关 键 词：非线性去噪 特征提取 图像分割 聚类分割算法 小波变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]