Image Analogies

(For more details look at Aaron Hertzman's Image Analogies web site at NYU)

We present a new framework for processing images by example, called "image analogies." Rather than attempting to program individual filters by hand, we attempt to automatically learn filters from training data. For example, the following figure demonstrates an image analogy used to learn a painting style:

'Unfiltered' painting (A) : 'Filtered' painting (A') :: Input image (B) : Target image (B')

The images on the left are training data; our system "learns" the transformation from A to A', and then applies that transformation to B to get B'. In other words, we compute B' to complete the analogy. (Only partial images are shown above; here are the full images).

Many examples and results are shown on these pages. For additional details of the algorithm, please see the paper.


We applied the image analogies approach to several different problems:

Other uses of our software:


Texture-by-numbers video 25MB, Running time: 2:08. (Note: Windows Media Player sometimes will only show the first minute if you directly play the movie; save the movie to your hard drive to avoid this problem.)


The Image Analogies software is available.

Paul Harrison's Resynthesizer GIMP plug-in does something similar, though the algorithm is different.


Image Analogies
A. Hertzmann, C. Jacobs, N. Oliver, B. Curless, D. Salesin.
SIGGRAPH 2001 Conference Proceedings.

Algorithms for Rendering in Artistic Styles
A. Hertzmann. Ph.D thesis. New York University. May, 2001.

Copyright © 2001 Aaron Hertzmann, Charles E. Jacobs, Nuria Oliver, Brian Curless, David H. Salesin