Physics-motivated features for distinguishing photographic images and computer graphics
Proceedings of the 13th annual ACM international conference on Multimedia
International Journal of Computer Vision
Efficient graphical models for processing images
CVPR'04 Proceedings of the 2004 IEEE computer society conference on Computer vision and pattern recognition
Automatic motion-guided video stylization and personalization
MM '11 Proceedings of the 19th ACM international conference on Multimedia
Oil painting rendering through virtual light effect and regional analysis
Proceedings of the 10th International Conference on Virtual Reality Continuum and Its Applications in Industry
| Hi-index | 0.00 |
An interesting and potentially useful vision/graphics task is torender an input image in an enhanced form or also in an unusualstyle; for example with increased sharpness or with some artisticqualities. In previous work [10, 5],researchers showed that byestimating the mapping from an input image to a registered(aligned) image of the same scene in a different style orresolution, the mapping could be used to render a new input imagein that style or resolution. Frequently a registered pair is notavailable, but instead the user may have only a source image of anunrelated scene that contains the desired style. In this case, thetask of inferring the output image is much more difficult since thealgorithm must both infer correspondences between features in theinput image and the source image, and infer the unknown mappingbetween the images. We describe a Bayesian technique for inferringthe most likely output image. The prior on the output image P ( X)is a patch-based Markov random field obtained from the sourceimage. The likelihood of the input P ( Y|X )is a Bayesian networkthat can represent different rendering styles. We describe acomputationally efficient, probabilistic inference and learningalgorithm for inferring the most likely output image and learningthe rendering style. We also show that current techniques for imagerestoration or reconstruction proposed in the vision literature(e.g., image super-resolution or de-noising) and image-basednon-photorealistic rendering could be seen as special cases of ourmodel. We demonstrate our technique using several tasks, includingrendering a photograph in the artistic style of an unrelated scene,de-noising, and texture transfer.