Noisy image super-resolution with sparse mixing estimators

Image super-resolution reconstruction (SR) has drawn a lot of attentions lately. But almost all existing SR algorithms do not consider about the noisy image SR problem. This paper proposes a novel super-resolution algorithm for noisy images based on sparse mixing estimators. Firstly, sparse mixing estimators are introduced to achieve a directional and sparse representation of noisy low resolution (LR) image. Then, we employ the median filter to define thresholds using the local characters of the sparse representation. After the noise is removed by shrinkage thresholds, the adaptive interpolations are adopted to achieve high resolution (HR) image. Experimental results demonstrate that our algorithm shows satisfactory performance in noisy image super-resolution reconstruction.