On the relevance of an algorithm to a problem

The increasing complexity of options available in image processing suites has not been met by a commensurate increase in the sophistication of algorithm selection mechanisms. So the user is forced to apply trial-and-error to discover the algorithm and the parameter settings which give the quality of output required for a particular application. The paper introduces the notion of the relevance of an algorithm to a problem, as a measure which selects algorithms appropriate for solving that problem. The operation considered is image registration. The author formally defines a problem in terms of (i) the data on which the operation is performed, (ii) the end application for which the operation is being performed. She defines algorithms in terms of their procedural steps and the constraints implicit in the assumptions of their development. She uses these concepts to define the relevance of an algorithm to a problem as the weighted sum of the number of common token types and properties of token types.