SUSAN Low Level Image Processing

Research by Stephen M. Smith.

SUSAN is an acronym for Smallest Univalue Segment Assimilating Nucleus. The SUSAN algorithms cover image noise filtering, edge finding and corner finding. For information on applying SUSAN nonlinear noise reduction to 3D images, see here.

Published Paper

The SUSAN principle is the basis for algorithms to perform edge detection, corner detection and structure-preserving image noise reduction. Papers on SUSAN have been published in BMVC92, ICPR96 and IJCV, and a patent has been granted. For a complete report on SUSAN either download the PostScript (susan.ps.gz - 1 Meg / susan.ps - 8 Megs) or view the html:

Low level image processing - title page

Note on Thinning

I have also written a short note describing the thinning algorithm used in SUSAN edge detection. Either download the PostScript (thinning.ps - 76K) or view the html.

Source Code

The source code for SUSAN is available in a self-contained C program which inputs and outputs PGM format images of any size, performing any of the three SUSAN algorithms. The source code is here. Instructions for compiling and running are included in the file. Please email me (steve@fmrib.ox.ac.uk) if you download the code so that I can send out program updates easily.

Note: the program is © Copyright 1995-1999 Defence and Evaluation Research Agency, UK. DERA has now become QinetiQ, who hold the patent on SUSAN. I'm sorry but I don't have any contact details for relevant people at QinetiQ, and am not aware of whether they have kept the patent current.

Test Image

You can also download my PGM format test image (test.pgm - 64K / test.pgm.gz - 1K). To find all features set t=10.