摘要
大部分的图像存在遮挡现象,因此只能获得对象的部分轮廓。曲率尺度空间描述子(Curvature scale space descriptor,CSSD)是MPEG-7标准采用的闭合轮廓描述子。本文研究并分析了闭合完整轮廓和部分开轮廓曲线演化过程的不同之处,提出了一种通用曲率尺度空间描述子(GCSSD),不仅继承了CSSD的旋转、尺度、平移不变性及抗噪能力,并较好地描述部分开轮廓的特征。本文也给出相应GCSSD的匹配算法,将其应用于彩色图像花的检索,实验结果表明其性能明显优于CSSD。
For most of images, the occlusion between objects is normal and common. So after preprocessing, only partial contour of an object from images is obtained. Human can recognize objects just by part shape of the object, good descriptor should perform retrieval of partial contours. MPEG-7 standard adopts curvature scale space descriptor （CSSD） as a close contour descriptor. By analyzing the difference between the whole close contour and partially open contour during the contour evolution a general curvature scale space descriptor （GCSSD） is proposed. The GCSSP can be invariant to rotation, scale, translation and robust to noise for inheriting CSSD and describe partially open contour. A relevant matching algorithm for GCSSD is used in flower retrieval. Experimental results show that GESSD performance is obviously better than that of CSSD.
出处
《数据采集与处理》
CSCD
北大核心
2007年第2期132-137,共6页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(60372059)资助项目
关键词
MPEG-7
空间描述子
图像检索
曲率尺度空间
部分开轮廓
MPEG-7
curvature scale space descriptor（CSSD）
image retrieval
curvature scale space
partially open contour
