Various analytical or diagnostic techniques (e.g., evaluation of 2-D electrophoretograms, discrimination, and counting of objects) deal with pictures containing many similar patterns or objects and a few which are dissimilar. In this paper an efficient method is described which analyzes such pictures. It proceeds in three stages: removal of film background, separation into regions enclosing a single object or pattern, and classification of these regions into groups containing objects that are similar. The classification is based on moment analysis of the separated regions. The limitations due to noise (film granularity and fluctuations in object shape) on the selection of moment orders are discussed. Using selected integral functions (moments) and a corresponding choice of variables for the representation of 2-D space, adapted to the particular characteristics of the patterns to be classified, leads to a fast method for discriminating patterns that are difficult to distinguish by visual inspection because of their low texture content. The intended application of the method is automated data extraction from 2-D electrophoretograms. The performance of the procedure is shown in two examples.
© 1984 Optical Society of America
Equations on this page are rendered with MathJax. Learn more.