Abstract: Nowadays，there is few methods for spherical panoramic mosaics which can cover all the visual range，and the existing ones have some limitations，such as instable iterative optimizations，and particular imaging equipments.In order to solve these problems，this paper presents a new approach to create spherical panoramas automatically，and the input images are just acquired by general equipments.This approach adopts the FFT-based phase-correlation algorithm to compute the transform relationships between images approximately.For avoiding the error accumulations during the stitching process，it creates a set of longitudinal-block images from the input images，which finally are stitched to a spherical panorama by the piecewise interpolation strategy.In many situations，the experimental results demonstrate that this approach has high efficiency，and gain high-quality mosaics.

Key words: spherical panoramas, Fast Fourier Transform（FFT）, piecewise cubic interpolation

摘要： 目前，能够覆盖所有视觉范围的球状全景图拼接方法较少，而且存在着一定的缺陷，例如迭代优化算法的不稳定性，图像采集设备的特殊性。为了解决这些问题，提出了一种自动的球状全景图拼接方法，它只需使用普通设备进行图像采集。该方法首先采用基于傅立叶变换的相位相关算法近似求解图像之间的变换关系。在拼接过程中，为了避免误差积累，首先生成一系列纵向分块图像，然后通过分段插值策略将它们拼接成无缝的球状全景图。通过对多种场景的实验表明，该方法计算效率高，并且能够达到满意的拼接效果。

关键词: 球状全景图, 快速傅立叶变换, 分段三次插值