Abstract: The invention in a preferred embodiment recognizes symbols by determining the center of mass and maximum extremity of a symbol candidate, and comparing it to a reference library after normalizing with respect to scale, orientation and center of mass.

Abstract: A method and apparatus is described in which a document is scanned and the coordinates of each graphic element intercepted along the scan line are analyzed to encode a compact digital representation of the various graphic elements. In one embodiment, detected features of a height less than a first dimension or having both a width less than a desired width threshold and a height less than a desired height threshold are deleted as representative of noise. No data representative of topological or dimensional features is output until after the continued recognition of such features for more than a threshold number of scan lines.

Abstract: An apparatus codes a scanned document locally representing each graphic element within a prescribed measure of accuracy by a trapezoidal approximation. The invention also includes a method for similarly coding a scanned document. The apparatus determines whether each scanned run is indicative of a Y- or .lambda.-junction, the termination of an old, or the commencement of a new graphic element, and whether a new linear approximation is necessary. The invention in a preferred embodiment recognizes symbols by determining the center of mass and maximum extremity of a symbol candidate, and comparing it to a reference library after normalizing with respect to scale, orientation and center of mass. In a preferred embodiment, an adaptive threshold parameter governs coding so as to reject noise and optimize a pair of linear predicters in a small number of scans. In a further preferred embodiment the accuracy of the linear predicters is refined so that the error is exponentially bounded.