Graphics & Imaging

Priors for Large Photo Collections and
What They Reveal About Cameras

Aseem Agarwala, Adobe Systems
Dan Goldman, Adobe Systems
Sujit Kuthirummal, Columbia University
Shree Nayar, Columbia University

A large photo collection downloaded from the internet spans a wide range of scenes, cameras, and photographers. In this project we discovered several statistics of such large photo collections that are independent of these factors. This makes it possible to recover the radiometric properties of a particular camera model by measuring how the statistics of images taken by one model differ from those of a broader photo collection with camera-specific effects removed. We show that using this approach we can recover both the non-linear response function of a particular camera model and the spatially-varying vignetting of several different lens settings. All this is achieved using publicly available photographs, without requiring acquisition of images under controlled conditions or even physical access to the cameras. We also applied this concept to identify bad pixels on the detectors of specific camera instances.

Publications

Kuthirummal, S., Agarwala, A., Goldman, D. B, Nayar, S.. 2008. Priors for large photo collections and what they reveal about cameras, In Proc. European Conference on Computer Vision, LNCS 5305, 74–87