| 摘要 |
Restoration based super-resolution combines a video sequence of low-resolution noisy blurred images using maximum likelihood estimation without regularization to produce a higher resolution image or video sequence up to a maximum enhancement factor r for the given video sequence. The maximum r value for a given sequence of low-resolution images that guarantees each high-resolution bin (pixel) will have at least one assigned pixel value from a low-resolution image is calculated. With the knowledge of max r, the system can select an r less than or equal to max r and thus operate without regularization. System robustness is further enhanced by providing a modified MLE estimator that can perform super resolution for arbitrary real-valued r so that enhancement is not limited to integer values.
|
