Newsgroups: comp.constraints
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!newsfeed.pitt.edu!dsinc!ub!acsu.buffalo.edu!rchopra
From: rchopra@acsu.buffalo.edu (Rajiv Chopra)
Subject: Expensive constraint checks
Message-ID: <D3sqwG.I5L@acsu.buffalo.edu>
Sender: nntp@acsu.buffalo.edu
Nntp-Posting-Host: diadem.cs.buffalo.edu
Organization: SUNY at Buffalo / Ctr for Document Analysis and Recog (CEDAR)
Date: Fri, 10 Feb 1995 18:19:27 GMT
Lines: 33


Hi!
	Has anyone addressed constraint satisfaction problems (assume 
binary, for now) where the constraints are not enumerated as lists of
acceptable tuples but implicit relations which might be expensive to verify?

For example, in an application of constraint satisfaction to computer vision, 
a constraint between two detectable objects A & B  (modelled as 
variables in the CSP) in the image might be: Object A is "brighter" 
than Object B. Verification of this constraint for potential labels (image 
segments) for A and B might involve image processing as opposed to a 
membership test in the corresponding constraint list.

Algorithms like AC-4 are optimal based on criteria like number of variables,
number of constraints. 

1) Is it also optimal in the number of constraint checks
required?
2) Are there other approaches to problem reduction which minimize the 
number of constraint checks required? (Assume the CSP is tightly constrained
and all solutions are desired) 
3) Like Ordering of Values and Variables in search, are there any heuristics
and/or results for "Constraint ordering" ?

Pardon my ignorance if these questions are naive. I might not have read 
enough of CSP literature.


Thanks in advance,

Rajiv Chopra
Graduate student, Dept. of CS, UB

