位置:成果数据库 > 期刊 > 期刊详情页
一种鲁棒的图像局部仿射不变特征提取方法
  • 期刊名称:模式识别与人工智能
  • 时间:0
  • 页码:-
  • 分类:TP391[自动化与计算机技术—计算机应用技术;自动化与计算机技术—计算机科学与技术]
  • 作者机构:[1]国防科技大学信息系统与管理学院,长沙410073
  • 相关基金:国家自然科学基金(60872150,60902091,61175006,61175015)资助项目
  • 相关项目:快速鲁棒的图像局部仿射不变特征提取方法
中文摘要:

为解决折反射全向成像分辨率低且分布不均匀问题,提出互补结构的全向成像方法,结合全向图内外环的互补特性进行全景图像融合及分辨率增强研究。其中,2幅配对全景图像之间的精确配准是必要前提和基础:首先根据反射镜边沿成像的椭圆边界线对反射镜空间姿态进行标定,基于标定结果将全向图的内环和外环分别展开为2幅校准后的柱面全景图像;然后基于普适最小二乘法求解多项式变换模型,对2幅初步配准的全景图像进行更精确的亚像素级配准。实验结果表明,算法满足图像融合的配准精度要求,并通过融合效果进一步验证算法的有效性。

英文摘要:

To resolve the problems of low resolution and non-uniform distribution in catadioptric omnidirectional ima- ging, a novel omnidirectional imaging method with complementary structure is proposed;and the studies on panoramic image fusion and resolution enhancement are carried out based on the complementary characteristics in the inner and outer rings of omnidirectional images. Notably, the precise registration of the two paired panoramic images is the es- sential precondition for further research. Firstly, the spaeial poses of the reflecting mirrors are calibrated according to the elliptical boundaries of the mirror boundary imaging, then the inner and the outer rings of the omnidirectional im- age are unwrapped into two calibrated cylindrical panoramic images respectively based on the calibration results. Then, the generalized least squares (GLS) algorithm is used to solve the polynomial transform model, and more accu- rate sub-pixel registration of the two primarily registrated panoramic images is performed. Experiment results demonstrate that the proposed algorithm fulfils the registration accuracy requirement of image fusion, and the fusion results further verify the effectiveness of the proposed algorithm.

同期刊论文项目
期刊论文 45 会议论文 5 著作 1
同项目期刊论文