一种基于Contourlet变换的多尺度纹理分割的新算法

NEW ALGORITHM OF MULTISCALE TEXTURE IMAGE SEGMENTATION BASED ON CONTOURLET TRANSFORM

受到基于模型的纹理分析方法的启发,提出了一种新的特征提取方法—有限混合纹理模式(Finite Texture Mixture Pattern,FTMP).FTMP是一个二元组的集合,可以通过聚类的方法进行计算.首先,基于Contourlet变换计算纹理的多尺度多方向变化信息;其次,对各尺度、各方向的变化信息分别进行聚类.这些聚类中心以及它们所占的比例组成的二元组的集合就构成了纹理图像的FTMP,反应了不同尺度不同方向的主要变化信息.这种纹理特征的计算方法充分利用了基于模型方法的基本思想,但却避免了复杂的参数计算.在FTMP的基础上,本文给出相应的纹理分割框架CFTMPseg,并通过定量和定性实验验证了所提算法的有效性.

A new feature extraction method, finite texture mixture pattern （FTMP）, was proposed inspired by the basic idea of pattern-based texture analysis methods. FTMP was a two-tuplet set, which could be obtained by clustering methods. Firstly, the multi-scale and multi-direction variations were calculated based on the Contourlet transform. Secondly, these variation information at different scale and in different direction were clustered into groups respectively. The centers and their corresponding proportion made up FTMP, which reflected the primary information＇s variations of different scales and different directions. Such a feature extraction method made full use of the idea of pattern-based method, but avoided the complicated parameter estimation and expression computation. Thus a supervised multi-scale texture image segmentation algorithm based on Contourlet transform-CFTMPseg was proposed based on FTMP, and its effectiveness was proven by quantitative and qualitative experiments.