Image Recovery by Simulated Annealing
Abstract
The basic problem of image recovery and pattern recognition is to determine the original pattern, f, given its corrupted version, g. The unknown pattern f is an element of the set S = f1, f2, f3, and the task is to deduce which pattern in S gave rise to the image data, g. S is called the solution candidate space and could be, for example, the set of alphabetical symbols. If it is known that certain elements of S have a higher probability of occurring than others (such as alphabetical symbols in text), this a priori information can be incorporated into the procedure for finding f according to the techniques of Bayesian analysis. In the general image recovery problem, S is the set of all possible patterns on an n x n pixel image, and the relationship between the original image, f, and the image data, g, can be modeled by g = f + 2, where w is random noise. Set of pixel brightnesses at lattice position (i,j) are described by f = fij, g = gij, and w = wij. Image recovery and pattern recognition problems are thus combinatorial optimization problems, in which a solution candidate space, S, must be searched. The larger S is, the more difficult the search.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1990
- Accession Number
- ADA230655
Entities
People
- David H. Gerstman
- James B. Cole
Organizations
- Harry Diamond Laboratories