A Conditional Random Field for Automatic Photo Editing
| Citation: | Brand, M.; Pletscher, P., "A Conditional Random Field for Automatic Photo Editing", IEEE Conference on Computer Vision and Pattern Recognition (CVPR), ISSN 1063-6919, pp. 1-7, June 2008 (IEEE Xplore) |
| MERL Report: | TR2008-035 |
| MERL Contact: | Matthew Brand |
We visually represent the distribution over labellings with a false-color image that indicates, at each pixel site, the label having maximum marginal likelihood (henceforth, maximum-of-marginals). Class likelihoods in turn inform local texture filtering or replacement, here yielding a resynthesized image with red-eye corrected and blemishes removed, but birthmarks preserved. The entire process is automatic and trained discriminatively from before-and-after images.
We introduce a method for fully automatic touch-up of face images by making inferences about the structure of the scene and undesirable textures in the image. A distribution over image segmentations and labelings is computed via a conditional random field; this distribution controls the application of various local image transforms to regions in the image. Parameters governing both the labeling and transforms are jointly optimized w.r.t. a training set of before-and-after example images. One major advantage of our formulation is the ability to marginalize over all possible labeling and thus exploit all the information in the distribution; this yield better results than MAP inference. We demonstrate with a system that is trained to correct red-eye, reduce specularities, and remove acne and other blemishes from faces, showing results with test images scavenged from acne-themed internet message boards.
 | Read the full technical report (PDF: 6 MB) |