SINGLE-FRAME SUPER-RESOLUTION BY INFERENCE FROM LEARNED FEATURES

Super-resolution is the creation of higher resolution views of pixel-based images through interpolation between the original pixels. Greater super-resolution can be achieved by taking advantage of local regularities inherent in natural images. In this paper, to learn regularities, we make use of the recently proposed SINBAD model  of how the cerebral cortical network learns regularities by discovering regularity-simplifying environmental features [5, 14]. Using the regularities discovered with the SINBAD approach, we were able to predict more accurately the interpolated pixels from the ones in the original image and were able to generate visually plausible fine spatial details in the expanded image