Implementation Schemes of Regularization Super-Resolution Image Reconstruction

This paper proposes two effective synchronous and parallel recursion schemes to implement regularization super-resolution image reconstruction. In the synchronous recursion, iteration step is adaptively adjusted by the speed of gradient descent to each observation channel. When blur support is too large or low-resolution images are severely degraded, however, the high-frequency information of the desired high-resolution (HR) image is still smoothed. So for fusing the information from different observation channels more effectively, parallel recursion is proposed to reconstruct desired HR image. In the two recursion schemes, spatial integration in down-sampling process is removed as well as system blurs, and nearest interpolation in up-sampling process is used to restrain edge artifact. Simulation results demonstrate that the two proposed implementation schemes give more satisfying results in both objective and subjective measurements.