基于隐马尔可夫复合树模型的图像纹理分析

提出了一种纹理图像隐马尔可夫捆绑树(HMT-b)模型的建模方法。该方法通过对小波分解后的三个子带(HH，HL，LH)中相应节点捆绑后作为一棵复合树进行建模，改进了迭代算法，所建模型能更好地描述三个子带问实际存在的小波系数相关性；对于每个尺度中的小波系数分布，HMT-b采用高斯混合分布来拟合。同时研究了尺度系数基于小波域泊松分布的统计建模方法。

Texture Image Analysis Based on Hidden Markov Binding Tree Model

Two novel modeling methods, called HMT b and pHMT b, for texture characterization and model in the wavelet domain are presented based on HMT. Two method have same characters i.e. they all capture statistical dependencies across DWT sub bands by using the graphical technique. Whereas, HMT b uses wavelet coefficients, belonged to HL, HH, LH sub bands, pHMT b uses scale coefficients in LL. HMT b takes a of Gaussian mixture model to describe wavelet coefficients is described by a peak and heavy tailed non Gaussian density. On another hand, probability density function describes scale coefficients characterized by Poisson mixture model, proposed in pHMT b. Experimental result show that both HMT b and pHMT b provide the higher percentage of correct classification over 94% than that of HMT, HMT 3s and WES, based on a set of 55 Brodatz texture.