Super-resolution methods comparison

Four different algorithms are compared:
  • Almalence's PhotoAcute
  • Robust super-resolution
  • Normalized convolution
  • Iterated back-projection
A summary table of algorithms characteristics:
PhotoAcute Iterated back-projection Robust super-resolution Normalized convolution
Non-Iterative Yes No No Yes
Tunable to the imaging device Yes No No No
Robust to Noise Yes No No Yes (with additional pass)
Speed Very fast, 100-400 mul/add per pixel Slow, more than 10,000 mul/add per pixel Slow, more than 12,000 mul/add per pixel Very slow (singular value decomposition at every pixel)

ISO 12233 test chart

Test conditions:

  • ISO 12233 Test Chart
  • Sigma SD10 camera with Foveon sensor
  • 4 frames

Test chart with crop areas marked



The test suite and ISO 12233 test pattern are from LCAV - Audiovisual communications labaratory.

Closeup of test chart cropped areas processed with different superresolution algorithms:

Original crops
PhotoAcute
Robust Super-resolution
Normalized Convolution
Iterated Back-Projection

Real scene (street)

Test conditions:
  • Outdoor street scene
  • Casio EX-F1 camera
  • Moving clouds
  • 6 images
Street scene with crop areas marked:




Closeup of street scene cropped areas processed with different superresolution algorithms:

Original crops
PhotoAcute
Robust Super-resolution
Normalized Convolution
Iterated Back-Projection


Animated comparison: move the mouse over the links below to view corresponding image (requires JavaScript):




References

Iterated back-projection. M. Irani and S. Peleg, Improving resolution by image registration, Graphical Models and Image Processing, 53:231-239, 1991.

Robust super-resolution. A. Zomet, A. Rav-Acha, and S. Peleg, Robust Super-Resolution, Proceedings international conference on computer vision and pattern recognition (CVPR), 2001.

Normalized convolution. Tuan Q. Pham, Lucas J. van Vliet and Klamer Schutte, Robust Fusion of Irregularly Sampled Data Using Adaptive Normalized Convolution, EURASIP Journal on Applied Signal Processing, Vol. 2006, Article ID 83268, 12 pages, 2006.
***