基于相位相关的柱面全景图像自动拼接算法  

Cylindrical Panoramic Image Automatic Mosaic Algorithm Based on Phase Correlation

作　　者：吴宪祥[1] 郭宝龙[1] 王娟[2] 

机构地区：[1]西安电子科技大学机电工程学院ICIE研究所,陕西西安710071 [2]空军工程大学电讯工程学院,陕西西安710077

出　　处：《光学学报》2009年第7期1824-1829,共6页Acta Optica Sinica

基　　金：国家863计划(2006AA012127);国家自然科学基金(60802077;60572152);教育部博士点基金 (200660701004)资助项目

摘　　要：提出了一种基于相位相关法的柱面全景图像自动拼接算法。首先推导了柱面投影公式、柱面平移量和平面平移量之间的对应公式;然后利用改进的相位相关法计算出归一化相位相关度的峰值及峰值点坐标,利用峰值作为相关度准则判断两幅图像是否存在重叠部分,利用最大相关度求交自动确定头尾图像,利用峰值点坐标判断两幅相邻图像的位置关系;最后将环绕拍摄的序列图像投影到柱面坐标系下,根据计算得到的变换参数进行图像拼接并融合得到柱面全景图像。实验结果表明,该算法能有效地实现顺序混乱的序列图像的自动排序和柱面全景图像拼接,具有较高的稳健性和拼接精度。An improved cylindrical panoramic image automatic mosaic algorithm based on phase correlation is proposed. Firstly, the cylindrical projecting formulae are given. Then, the normalized cross-correlation function is computed, and the peak value and position of the normalized cross-correlation function are found. The peak value is used to estimate whether two images are overlapped. The first and the last image can be ascertained automatically by maximal correlation degree intersecting, and the relation of the overlapped images can be ascertained by the peak position. Finally, the rounding planar image sequences are projected to cylindrical coordinate, and the cylindrical panoramic image can be gained by stitching together the images in sequence with the calculated transformation parameters. Experimental results show that the proposed method outperforms traditional methods with high robustness and precision, which can sort the unordered image sequences and stitch the images in sequence into a cylindrical panoramic image effectively and automatically.

关 键 词：图像处理 图像拼接 相位相关 自动排序 柱面投影 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]