THE "CREEP-AND-MERGE" SEGMENTATION SYSTEM
Antranig Basman, Joan Lasenby and Roberto Cipolla
July 1997
We present a novel architecture for region-based segmentation of stationary and quasi-stationary statistics, which is designed to function correctly under the widest range of conditions. It is robust to the extremes of region topology and connectivity, and automatically maintains region boundaries sampled to the minimum scale at which the region configuration can be determined with statistical confidence. The algorithm is deterministic, and when operating on images from within its domain of validity, contains no adjustable parameters. In contrast to most other techniques directed at the same problem, the progress of the algorithm cannot be described by the optimisation of a global energy criterion.
We initially describe the system in an implementation--neutral form, and discuss the properties required of a statistical engine suitable for connection to it - the system is modular, and allows for straightforward replacement of the driving statistics by a wide variety of alternatives.
A specific implementation for 2-dimensional images is then presented and illustrated. Quantitative evaluation shows superior results to a collection of other systems. We then illustrate a successful application of the system, together with higher-level processing, to the difficult problem of segmentation of medical images gathered by ultrasound.
[1.3 MBytes compressed PostScript, 32 pages]
If you have difficulty viewing files that end '.gz',
which are gzip compressed, then you may be able to find
tools to uncompress them at the gzip
web site.
If you have difficulty viewing files that are in PostScript, (ending
'.ps' or '.ps.gz'), then you may be able to
find tools to view them at
the gsview
web site.
We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.