Neighborhood regularized image superresolution for applications to microscopic imaging

We aim to computationally improve the resolution of images acquired from low magnification slide scanners. Such images, while efficient in terms of storage and data transfer, do not have the necessary resolution needed in diagnostic settings. The question we ask is, can the high resolution image be reconstructed, given a low resolution counterpart. This problem, known as the image superresolution problem (SR), is widely studied in the context of natural images (obtained from digital cameras). But for the application at hand, there are unique challenges, since the images we use do not have the high level features which are typically used in generating superresolution reconstructions. To address these issues, we design an approach specifically tailored for our application. We develop a deep learning framework that implements regularized sparse coding to smoothly reconstruct high-resolution images, given their low-resolution counterpart. Results show that our method indeed produces images which are similar to images from high resolution scanners, both in quality and quantitative measures and compares favorably to several state-of-the-art methods across a number of test images.