Automated Pictorial Pattern Analysis - Programming Methods

Abstract

Progress on programming techniques for automated analysis of three-dimensional matrices representing cell clusters is reviewed. Our philosophy for picture processing is based upon the unproved assumption that an empirical relation exists which will allow positive identification of a useful fraction of a given cell type. This means that we are willing to accept a program which will identify only a fraction of the cells, provided such identification is very reliable. At present, we are working with 32 x 32 element matrices containing light intensity values. The picture analysis is carried out by an IBM 360/50. Tested programs include those which will allow for illumination correction, sharpening, intensity contours leading to data representation in a binary matrix form. The methods employed are discussed and illustrated.