Image Analogies

Overview

Image Analogies is a framework for processing images by example. Rather than programming a separate filter for each effect, the system receives a source image A, its filtered version A', and a new input B. It searches the training pair for local image neighborhoods that match B and synthesizes B' so that the relationship between B and B' resembles the relationship between A and A'. A coarse-to-fine image pyramid and a coherence term help neighboring output pixels draw from consistent parts of the example.

Applications

• Toy filters — blurring, embossing, and other simple effects
• Texture synthesis — generating new textures from a single example
• Super-resolution — inferring a high-resolution image from a low-resolution source
• Texture transfer — applying arbitrary source textures to target images
• Artistic filters — synthesizing oil, pastel, and pen-and-ink rendering styles from scanned real-world examples
• Texture-by-numbers — creating realistic scenes using a simple painting interface

The paper demonstrated one framework across all of these visually different tasks, rather than evaluating a single effect with a numerical benchmark. Results depended strongly on the example pair and on local appearance: a transformation absent from the example could not be learned, and local matches could produce artifacts when an image required long-range or semantic consistency. The search and synthesis process was also computationally expensive.

Publications

@inproceedings{hertzmann2001image,
  author = {Aaron Hertzmann and Charles E. Jacobs and Nuria Oliver and Brian Curless and David H. Salesin},
  title = {Image Analogies},
  booktitle = {Proceedings of ACM SIGGRAPH 2001},
  pages = {327--340},
  year = 2001,
  doi = {10.1145/383259.383295},
  cites = 2426,
  citesdate = {2026-03-23}
}

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

Videos

• (6.2 MB, 2:08)