论文部分内容阅读
论述一种从图像序列中重建出全方位全景图像的技术 ,旨在克服镜头只是纯粹的摇摄时许多成像条件限制。在块匹配算法中 ,如何选取最佳配准基本块对于增强算法的鲁棒性和性能是非常重要的。本文在包含主要视觉特征的高频图像中自动选取最佳配准基本块 ,为了能够减少累积的配准误差 ,采用了基于相位相关的、带旋转校正的全局配准算法 ,从而得到解决了平移和旋转的优化图像块。然后 ,采用了基于 Levenberg-Marquardt非线性递归型最优化算法 ,解决由于镜头的视差和不规则变形带来的局部误差。文中还采用平滑滤波器解决了在两幅图像镶嵌出可能由于累积配准误差而出现明显的条缝
Discusses a technique for reconstructing omni-directional panoramic images from image sequences, designed to overcome the limitations of many imaging conditions when the lens is purely panning. In the block matching algorithm, how to choose the best registration basic block is very important to enhance the robustness and performance of the algorithm. In this paper, the optimal registration basic block is automatically selected in the high-frequency image containing the main visual features. In order to reduce the accumulated registration error, a phase-based global registration algorithm with rotation correction is adopted, And rotate the optimized image block. Then, based on Levenberg-Marquardt nonlinear recursive optimization algorithm, the local error caused by the lens’s parallax and irregular deformation is solved. The article also uses a smoothing filter to solve the two images inlay out may be due to the accumulation of registration errors and obvious slits