重复结构检测的建筑物立面影像纹理分割  

Image texture segmentation of building fa?ade based on repeated structure detection

作　　者：辛卫东[1] 刘亚文[1] XIN Weidong, LIU Yawen(Wuhan University, Wuhan 430079, Chin)

机构地区：[1]武汉大学,武汉430079

出　　处：《测绘科学》2018年第3期82-86,116,共6页Science of Surveying and Mapping

基　　金：中央高校基本科研业务费专项基金项目(2042014kf0294)

摘　　要：针对检测影像重复结构的算法需要大量迭代计算、效率不高等问题,该文结合相位相关和灰度相关算法,实现了建筑物立面影像纹理重复结构的检测,在此基础上,利用灰度梯度信息进一步分割立面纹理结构。相位相关通过在频域检测立面两区域影像傅里叶变换的相移来确定其在空间域对应的近似偏移量,实现重复纹理结构的粗定位。基于灰度相关的相关系数用来检测两区域影像的灰度相似性以精确确定重复纹理结构。纹理结构的分割则是根据纹理边缘梯度特点,在水平和竖直方向定位纹理结构单元的分割线。实验证明,该算法结合了立面影像纹理相似区域粗定位和精检测,有效提高了立面影像纹理重复结构检测及分割的效率。Building facade structure generally meets certain arrangement rules.For example,door,window and balcony are repetitive and symmetric.This paper presented a method of texture segmentation of building fa？ade based on repeated structure detection.It is further used for regularity extraction of fa？ade structure and fa？ade model reconstruction.In this paper,the repeated structure of building facade is first detected using phase correlation and gray correlation algorithms.Then the texture segmentation of building facade is further realized using split line evaluation function based on gray gradient.Phase correlation determines the approximate offset in spatial domain and the rough location of the repetitive texture structure by the way of phase-shift detection with Fourier transform in frequency domain.Correlation coefficient based on gray correlation is used to detect the gray similarity of two regions to accurately determine the repeated texture.Texture segmentation is based on the characteristics of the gradient of texture edges,and locates the division lines of texture structures in the horizontal and vertical orientation.Experiments show that the efficiency of repeated structure detection and texture segmentation is improved using rough positioning and precision detection of image similarity regions.

关 键 词：建筑物立面 相似性检测 纹理分割 

分 类 号：P232[天文地球—摄影测量与遥感]