A generalization of template matching for recognition of real objects

摘要

Pattern-recognition methods are discussed under the general headings of feature extraction and template matching. Problems arising from these methods are of two types: those arising in the process of recognition and those arising when making measurements on objects after recognition. It is indicated how template matching can avoid these latter types of problems and how a generalization of the normal notion of template can surmount problems of the first type. Such a generalized template matching procedure for recognizing cells in a microscope slide is discussed. The generalized template, in this case, makes use of the fact that the cells seem “bubble-like”. It expands a “bubble” template and detects matches between this and the visual data.