Abstract
This chapter presents a general framework for image super-resolution based on wavelet fusion is presented through discussion of three non-iterative algorithms for image super-resolution: linear minimum mean square error super-resolution, maximum entropy super-resolution, and regularized super-resolution using wavelet fusion. In super-resolution image reconstruction algorithms, several degraded low-resolution (LR) observations are used to estimate a single high-resolution image. The multi-channel image restoration step aims at obtaining multiple undegraded images of the same dimensions as those of the available LR images. Image restoration on a multi-channel basis aims at incorporating all the information existing in all the channels into the restoration process instead of restoring each channel separately. The super-resolution image reconstruction algorithms were tested on different noisy degraded LR observations with different signal-tonoise ratios (SNRs). The evaluation of the SNR in the available degraded images is another important issue that is closely related to the noise variance estimation from the degraded image.