中文摘要：

提出一种由图像序列来构造和调整全景图的新方法。首先利用基于Fourier-Mellin变换的相位相关算法进行相邻图像的初匹配;然后在迭代优化算法中加入光照变化模型,使用一种新的重叠区域亮度方差和公式来求解透视投影参数,进一步提高局部配准精度;最后,为满足整幅全景图的首尾一致性,我们提出一种新的全局调整策略来减少积累误差的影响,生成完整的360度柱面全景图。与现有其它算法相比,在手持相机无法精确控制其运动以及相邻照片存在亮度差别等情况下,使用该算法可以有效提高自动配准与拼接的成功率。实验结果表明该方法具有较高的准确度和较强的实用价值。

英文摘要：

A new method of construction and refinement of panoramic mosaics from sequences of images was proposed.First,the initial registration was taken with the phase correlation based on Fourier-Mellin transforms that enhanced the robustness of registration.Second,the registration precision was improved by calculating the parameters of the projection transform with an iterative optimization that includes additional illumination model.Lastly,to satisfy the head-tail consistence of panorama,a new global alignment method was proposed that reduced the accumulated error to create 360° full-view panoramic mosaics.Compared with other algorithms,this method can improve the success rate of the automatic registration and mosaic,especially when the hand-held camera had an irregular movement or adjacent images had difference of brightness.The experimental results showed that it had higher accuracy and fairly good practical value.