Efficient Super-Resolution Reconstruction for Translational Motion using a Near Least Squares Resampling Method

In this paper we propose a computationally efficient method for image super-resolution reconstruction. We concentrate on pure translational motion and shift-invariant blur. This is the case of interlaced sampling where each low resolution image is uniformly sampled but the overall sampling is non-uniform with respect to the high resolution grid. The reconstruction problem is considered in a multi-resolution framework using separable B-spline wavelets as basis. This allows performing the reconstruction on a uniform higher resolution grid by digital filtering. A computationally efficient structure-the transposed modified Farrow structure is used to achieve a near least squares solution without using any matrix inversions or iterations. The reconstruction can also be performed at non-dyadic scales. Our method minimizes the aliasing distortions and gives results comparable with other least squares techniques.