摘要
提出了一种新的自动全景图象拼合的鲁棒性算法。一是,基于多分辨率样条提出了一种新的图象拼接算法,它采用多分辨率样条来确定拼接区域宽度T和加权函数,使图象在不同分辨率下分别拼合,这样使拼合成的图象既清晰又光滑无缝;二是,在Levenberg-Marquardt非线性最小迭代算法的基础上提出了一种新的非线性最小迭代算法。算法在大多数情况下能自动完成,适宜于任何大小的图象并且不需要知道任何相机参数;对相邻帧之间相机运动没有严格的限制。
A new robustness algorithm is presented for automated panoramic mosaicing based on multiresolution spline technique. First, an iterative nonlinear minimization algorithm is presented based the Levenberg-Marquardt algorithm; Second, an image registration algorithm is presented based on multiresolution spline technique; in the procedure, the images are blended with different weighted functions in each frequency band. The coefficients of the weight function and the length of the transition zone are decided by the difference scale of the two images to be stitched in each frequency band. The algorithm can be finished with automating in most cases and suit for any scale image with any unknown parameter of camera (such for focal length).
出处
《系统仿真学报》
CAS
CSCD
2003年第6期895-897,共3页
Journal of System Simulation
基金
陕西省自然科学基金(2001X24)
关键词
绘制
虚拟现实
图象拼合
多分辨率样条
image based rendering
virtual reality
image mosaic
multiresolution spline
