Abstract: A character feature detection system is provided for reliably reading non-touching hand printed and multi-font machine printed alphanumeric characters. An information field is scanned by a sensor array, and the sensor responses are digitized to form a binary character image. The image is thinned in one-cell layers as the image boundary and loops internal to the image are traced from boundary point to boundary point. A transition from one boundary point to another is defined by one of eight equally divergent vectors which are indicative of the local slope of a boundary between connected points. The sequence of vectors and their x-ray locations are recorded, and boundary slopes are indicated by a running average of four vectors representative of cell to cell transitions. Character features are detected from the sequence of boundary slopes to form a feature set including circumference counts, maximum and minimum image boundary limits, inflection points and loops.