基于CHMT/NN的小波域纹理图象分割新算法  

Texture Image Segmentation Based on CHMT and NN in Wavelet Domain

作　　者：李会方[1] 徐瑞萍[1] 庞文俊[1] 

机构地区：[1]西北工业大学电子信息学院,西安710072

出　　处：《弹箭与制导学报》2005年第SC期729-732,共4页Journal of Projectiles,Rockets,Missiles and Guidance

摘　　要：文中提出了一种基于隐Markov模型和多层感知器的小波域图象纹理分割新算法。首先该算法通过图形组合方法有效地提取了图像在小波变换域各子带之间的相关性,然后应用多层感知器进行分类,将HMM的规范性和MLP神经网络的分类能力有效地结合起来。最后给出了文中算法对Brodatz纹理的分类结果。实验证明了文中算法的有效性。A new approach for texture Image Segmentation based on CHMT and MLP neural network in wavelet domain is presented.First the proposed algorithm is developed by using the graphical grouping technique to capture the subband dependencies of the texture DWT,and then used with a multilayer perception NN classifier The algorithm combines the time normalization property of the HMM with the superior discriminative ability of the neural net(NN).Some experi- mental results using Brodatz texture with integrated hybrid CHMM/NN classifier schemes are presented,and the re- sults are quite satisfied.

关 键 词：图象分割 小波 隐 MARKOV 模型 多层感知器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]