Super-Resolution Using a Markov Network Approach
One would like to have an intelligent method for expanding the resolution of an image. It should keep edges, which are implicitly described, in the low-resolution image sharp. It should make intelligent guesses about the details of textures. The focus of our recent work has been on simplifying the training and set-up computations, and reducing artifacts.
Background & Objective: Images are typically represented as collections of pixels, yet we would like to treat them as if they were resolution independent objects. In graphics, we have that option with polygon based representations: if one zooms in on a polygon-defined edge, the edge will stay sharp through all levels of zooming. We would like to have a similar level of resolution independence even in our pixel-based image representations.
Technical Discussion: We use a training based approach. We examine many pairs of high resolution, and low resolution versions of the same image data. We divide each image into patches, both high resolution and low-resolution patches. We form a training database of 50,000 - 100,000 high and low-resolution patches. Given a new low-resolution image, we seek to estimate the most probable corresponding high-resolution image. In the training database, there may be several different examples similar to any given input low-resolution patch. Typically, we gather a collection of 10 candidate high-resolution image patches to explain each low resolution input patch. The requirement of compatibility between the candidates of neighboring patches is used to select which of each patches 10 candidates is the best choice. We have improved the method used to measure compatibility between neighboring high-resolution candidate patches, which speeds up the set-up time needed to process an image.
| Technical Reports: | |
| Example-based super-resolution | |
Technology Areas:
Computer Vision
Artificial Intelligence
Graphics
Modification Date: January 23, 2007