Research on Computer-Assisted Generation of Image Processing Algorithms

To develop an algorithm for an image processing task a person normally should have both extensive experience in image processing and detailed knowledge of the task. Frequently those who need such algorithms have the detailed problem knowledge but not the image processing experience. We are investigating the extent to which the necessary emrience can be incornrated into .-a system for generating image processing algorithms. Such a system requires at least three types of knowledge: algorithm knowledge, operation knowledge, and selection rules for algorithms and operations. We represent the algorithm knowledge as a series of candidate algorithm architectures where each architechue is applicable to a range of tasks. The architectures consist of sequences of pseudo-commands specifying the main property required at each step of the algorithm. Operation knowledge includes 1) the primary property, preconditions, and potential side effects of the operation, 2) how the operation may be invoked and what parameters are required, and 3) how the results of the operation should be evaluated. The selection rules determine how the above two types of knowledge are combined to give a useful algorithm.