We envision a new generation of upsampling algorithms that draw upon the many images publicly available over the Internet. Consider the hypothetical photo enlargement system depicted above. A user enters an image they wish to “intelligently upsample”. While we would not expect to find a higher-resolution version of this exact same image on the Internet, we are very likely to find many with very similar components captured at higher resolutions. We are developing a system that evaluates 50 million on-line images, collects those with similar yet higher-resolution parts, and uses them to synthesize an enhanced version of the input. In the example above, such a system would select images that contain similar colored and shaped eyes, similar fur texture, and similar foliage. This project will study the appropriate theoretical framework, search algorithms, and data collection and processing techniques for example-based image super-resolution at massive Internet scales.

This website will serve as a repository for publications, datasets and software as they become available.

This project is partially funded by the National Science Foundation (IIS-084416) through their Cluster Exploratory (CluE) program and supported by a computer cluster maintained by Google and IBM.