多分辨模型下的无监督统计纹理分割算法.pdf

第 26卷第 1期 计算机应用 V ol. 26 No. 1 2006年 1月 ComputerApplications Jan. 2006 : 1001- 9081( 2006) 01- 0129- 03 周 晖, 王润生 ( 国防科学技术大学 ATR重点实验室, 湖南长沙 410073) (zhouhui_jj@ 163. com) : 针对基于统计的纹理分 算法存在的不足, 提出了一种新的多分辨模型下的无监督统计纹 理分 算法该算法分层次对纹理图像进行动态分 , 有效地利用了不同分辨率上最能表述某一纹理 特性的统计特征, 然后结合纹理的结构信息对边缘区域进行边界提取, 进而得到较准确的纹理分 图 实验证明了这种区域与边界相结合的算法对纹理图像的分 具有高效准确通用性强的特点 : 纹理分 ; 多分辨率; 动态聚类; 边缘提取; 无监督 : TP391. 41 : A Un supervised statistical texture segm en tation based on m u ltiresolu tion m odel HOU Hui, WANG Runsheng (A TR S tateK ey Lab, N ational University of D ef ense Techno logy, Chang sha H unan 410073, China ) A bstract: In order to overcom e the w eakness of statistical texture segm entation m ethod, a new unsupervised texture segm ent algorithm, based on m ultiresolution statisticalm ode,l was present. It first found the best feature that exp lained the texture clearly in d ifferent resolution and then the dynam ic cluster on different level. A t last, by comb in ing the structure inform ation of texture edge, the edge of d ifferent patterns w as extracted to get a relatively accurate result of texture segm entation. The results show that this algorithm based on region and edge is an efficient, accurate and robust m ethod of texture segm entation. K ey w ords: texture segm entation; mu lti resolution; dynam ic clustering; edge extracting; unsupervised 0 引言 , , [ 1] , , , , , , , , , ,