摘要
基于对偶树复小波(DT-CWT)和马尔可夫随机场(MRF)模型提出了一种监督纹理图像分割算法,算法包括两个步骤,首先对复小波变换系数进行较为精确的建模,提取其一阶统计信息作为纹理特征,综合多个尺度的信息,基于极大似然标准进行初始分割;其次,将初始分割结果用MRF模型表示,基于贝叶斯最大后验(MAP)融合初始分割结果,得到最终的分割结果。算法应用于合成纹理图像和实际图像得到了良好的结果,对比实验表明算法所采用的纹理特征的提取方法、小波变换方式、用MRF模型来建模标号等是算法简洁有效的基础。
A supervised texture image segmentation algorithm based on dual tree complex wavelet transform （DT- CWT）and Markov random field （MRF）model is proposed. The algorithm consists of two steps. First, the complex transform coefficients are statistically modelled; the first order statistics are extracted from the model as texture features. Initial segmentation class labels are obtained based on maximum likelihood criterion integrating several scales in formation. Then, the initial results are modelled by MRF, and final segmentation results are gained via Bayesian MAY estimation. It is supervised segmentation, the features for every class are estimated using the observed images in ad vance. Better segmentation results are obtained for synthesized Brodatz texture images and synthetic aperture radar images. The method owes its succinctness and effectiveness to several aspects, such as wavelet transform mode adopted, texture feature extracting method, class labels＇ further fusing based on MRF model, etc.
出处
《计算机科学》
CSCD
北大核心
2007年第1期187-190,共4页
Computer Science
基金
国家自然科学基金项目(60133010)
陕西师范大学校重点项目的资助
关键词
对偶树复小波变换
纹理图像分割
马尔可夫随机场
贝叶斯估计
Dual tree complex wavelet transform, Texture image segmentation, Markov random field, Bayesian estimation
