Patent US 4,048,615

A reliable optical character recognition system substantially insensitive to variations in character stroke size is provided for reading numeric and special characters printed on a document surface in plural fonts of unknown order. More particularly, a character is scanned and its image stored in a two-dimensional mosaic without requiring precise positioning therein, amplitude correlation for video signal quantization, or normalization of character stroke sizes. The image mosaic is scanned by a row-column analyzer to detect primary features of the image including the number of character segments in the rows and columns, the segment length and rightmost segment coordinate for those rows containing only a single segment, and the segment length for those columns having a single segment. Feature decision logic analyzes the primary features to form a set of secondary features based upon the occurrence and order of the primary features. The set of secondary features then are analyzed in a sequential logic decision tree to identify the character.