Contourlet HMT model with directional feature  

作　　者：WANG XiangHai CHEN MingYing SONG ChuanMing XU MengChun FANG LingLing 

机构地区：[1]College of Computer and Information Technology,Liaoning Normal University,Dalian 116029,China [2]State Key Laboratory for Novel Software Technology,Nanjing University,Nanjing 210093,China [3]Provincial Key Laboratory for Computer Information Processing Technology,Soochow University,Suzhou 215006,China

出　　处：《Science China(Information Sciences)》2012年第7期1563-1578,共16页中国科学（信息科学）（英文版）

基　　金：supported by National Natural Science Foundation of China (Grant No. 60372071);Natural Science Foundation of Liaoning Province (Grant No. 20102123);Liaoning BaiQian Wan Talents Program (Grant No.2008921036);the Open Subject of State Key Laboratory for Novel Software Technology,Nanjing University(Grant No. KFKT2011B09)

摘　　要：According to Shanno＇s information theory, the directional feature of texture is defined as the value of directional variable when an image signal attains a singularity of random distribution. In terms of this definition, we calculate the texture＇s directional features using Tamura＇s method and study the directional probability distribution of Contourlet coefficients. Then we find that the directional features tend to be conveyed across parent and child subbands. Based on this conclusion, we establish a novel probability distribution model of hidden direction variables under the condition of hidden state variable＇s distribution, named Contourlet HMT model with directional feature. The structure and training method of the model are presented as well. Moreover, an unsupervised context-based image segmentation algorithm is proposed on the basis of the proposed model. Its effectiveness is verified via extensive experiments carried out on several synthesized images and remote sensing images.According to Shanno＇s information theory, the directional feature of texture is defined as the value of directional variable when an image signal attains a singularity of random distribution. In terms of this definition, we calculate the texture＇s directional features using Tamura＇s method and study the directional probability distribution of Contourlet coefficients. Then we find that the directional features tend to be conveyed across parent and child subbands. Based on this conclusion, we establish a novel probability distribution model of hidden direction variables under the condition of hidden state variable＇s distribution, named Contourlet HMT model with directional feature. The structure and training method of the model are presented as well. Moreover, an unsupervised context-based image segmentation algorithm is proposed on the basis of the proposed model. Its effectiveness is verified via extensive experiments carried out on several synthesized images and remote sensing images.

关 键 词：directional feature Contourlet HMT Contourlet HMT with directional feature unsupervised texture segmentation 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TB114.3[自动化与计算机技术—计算机科学与技术]