Super-resolution reconstruction is an effective method to solve the problem that the face image resolution is too low to recognition in video, but the non-rigid change of face distortion and expression changes greatly affect the accuracy of registration and reconstruction. To solve these problems, a method of multi-level model Free Form Deformation (FFD) elastic registration algorithm based on B spline is proposed. It first uses low-resolution FFD grid for global registration, then divides the global registration image into several sub-images pairs and calculates the correlation coefficient of each pairs. At the same time, local registration with high-resolution FFD grid is done when the correlation coefficient of sub-image is small. In the process of optimization, adaptive step length gradient descent algorithm based on chaotic variables is used to improve optimization efficiency. After registration, the algorithm of POCS is used to reconstruct high resolution image through several low resolution images to realize recognition. Experimental results from standard video database and self-built video database show that this method can register and reconstruct accurately in the condition of great facial distortion and expression variation, and the face recognition accuracy is also improved.