Global motions in a video sequence caused by camera motion are often modeled by parametric transformations (motions) of two-dimensional images. The process of estimating the motions parameters is called global motion estimation. Global motion estimation is a useful tool widely employed in computer vision, video processing, and other applications. In this work, we focus on speeding up the global motion estimation in the framework of global motion models defined in the MPEG-4 video coding standard. In MPEG-4, the Levenburg-Marquardt algorithm (LMA) is performed iteratively to minimize a nonlinear objective function in estimating the global motion parameters. The minimization process is expensive computationally due to the involvement of all the pixels within an image frame. We propose to reduce the computation complexity by using only part of the image data in two stages of the LMA. The first stage is to obtain a good initial guess of the transformation parameters, which is critical to the final convergence of the algorithm. While the initial guess is chosen based on the output of a three-step search in MPEG-4, we propose a new method for determining the initial guess by applying the LMA itself on a very small portion of the pixels. The complexity of computing the initial guess can be lowered by using just a small number of iterations. The second stage of the LMA is to produce the final motion parameters in an iterative fashion, based on the coarse estimate of the motion parameters obtained in the previous stage. The LMA in this stage again operates on a subset of the pixels to further reduce the computational complexity.
展开▼
