Abstract: A character segmentation process for skewed documents divides the document bit map into vertical blocks and within each vertical block estimates initial top and bottom bounds of each character in a character row from horizontal projections of the vertical block, thereby avoiding the effects of skew. Operations in the document bit map to compute the exact top, bottom, left and right boundaries of the character are confined within the narrow strip between the initial top and bottom bounds of the character so as to greatly reduce the amount of data fetched from memory during such operations and thereby speed up the segmentation process.

Abstract: During segmentation of individual characters in a column, the invention predicts for the next character an optimal minimum gray level sensitivity threshold while scanning pixel rows between characters by lowering the threshold unless or until noise in the image causes the scanner to detect too many "ON" pixels between adjacent numerals. During subsequent attempts to recognize the segmented character, the invention computes a confidence score and increases the sensitivity threshold above the predicted level if the confidence score is insufficient.

Abstract: A feature-based character recognition identification and confidence level are determined for an unknown symbol. If the confidence level is within an intermediate range, the feature-based identification is confirmed by matching the unknown character with a reference template corresponding to the feature-based identification. If the confidence level is below the intermediate range, template matching character recognition is substituted in place of the feature-based identification. If the template matching recognition identifies more than one symbol, corresponding templates from a second set of templates having thicker character strokes are employed to resolve the ambiguity.

Abstract: The width of each stroke in a character image is reduced one pixel at a time over a series of repetitive cycles until the number of cycles exceeds a predetermined number proportional to the thinnest character stroke observed in the character image. The process is then continued until the character has been skeletonized, but with the added constraint that the existing end point pixels of character strokes are retained.

Abstract: Character images which are to be sent to a neural network trained to recognize a predetermined set of symbols are first processed by an optical character recognition pre-processor which normalizes the character images. The output of the neural network is processed by an optical character recognition post-processor. The post-processor corrects erroneous symbol identifications made by the neural network. The post-processor identifies special symbols and symbol cases not identifiable by the neural network following character normalization. For characters identified by the neural network with low scores, the post-processor attempts to find and separate adjacent characters which are kerned and characters which are touching. The touching characters are separated in one of nine successively initiated processes depending upon the geometric parameters of the image.