摘要
提出了一种提取户外建筑目标图像中布局信息的方法.首先,基于超像素技术对所给的图像进行大致区域划分.超像素技术是基于一个测度谓词,其利用图像的基于图论的表示法来判定两区域的边界;其次,以划分后的区域(称为超像素)为单位,利用颜色、位置、纹理等信息对其进行标记.在标记纹理特征时,采用了基于3D基元的纹理识别方法.最后,定义规则整合各项标记,实现了对图像内容的划分,提取其布局信息.实验结果表明,该方法应用于常见几种布局的户外建筑目标图像都能收到较好的效果.
A algorithm was proposed to grasp the rough surface layout of a outdoor building scene. It took the first steps towards segmenting an outdoor building image into regions using superpixel-based method, This mothod was based on a predicate which uses a graph-based representation of the image to measure the evidence for a boundary between two regions. Secondly,it used all of the available cues: material,location,texture gradients etc. to label the regions. A texture recognition algorithm based on 3D texton was used to capture the texture features. Finally,it combined all of the cues in order to labeling regions of the input image into coarse categories:"ground","sky",and "vertical". The results indicate that our method is able to create beautiful models for several kinds of common layout of outdoor building images.
出处
《厦门大学学报(自然科学版)》
CAS
CSCD
北大核心
2010年第2期175-180,共6页
Journal of Xiamen University:Natural Science
基金
国家重点基础研究发展计划(973计划)项目(2007CB311005)
福建省自然科学基金计划资助项目(A0710020)
厦门大学985二期信息创新平台项目
关键词
超像素
图象分割
特征提取
3D基元
superpixel
image segmentation
~eature extraction
3D texton