Image mosaicing is widely used in computer vision applications. In this paper, a new globally consistent image alignment method for video mosaicing is presented. Due to various uncertainties on noise, illumination, and modeling, the problem of global image alignment is considered as a stochastic estimation problem. The augmented system state consists of the transformation parameters of video image sequences. The system motion model, the augmentation model, and the observation model are constructed. The homography parameters of image sequences are augmented and estimated recursively with augmented Kalman filter in a common state vector and covariance matrix. Some experimental results are provided to validate the performance of the proposed method. The proposed image alignment method can handle the uncertainty efficiently. Image alignment is globally consistent and accurate.