Character recognition system and method using spatial and structural feature extraction

A character recognition system comprising:

• a data storage device;
• image acquisition apparatus configured to capture an image of an object character and store the character image in the data storage device;
• feature extraction apparatus communicatively linked to the data storage device and configured to receive the character image and (i) algorithmically scan the character image along each scan angle of a predetermined set of scan angles in order to extract character features including bars, lakes and bays ascertainable along that scan angle and (ii) assemble a feature vector corresponding to the character image, the feature vector including data indicative of the character feature types present in the character image along each scan angle;
• and
• character recognition apparatus configured to receive an assembled character feature vector from the feature extraction unit and recognize a character corresponding to the assembled feature vector at least partially on the basis of the quantity of each of (i) bars;
• (ii) lakes and (iii) bays indicated in the feature vector,
• wherein the character image comprises character image pixels including character pixels and contrasting background pixels and the feature extraction apparatus is programmed to:
• (i) scan the character image along a plurality of parallel scan lines oriented at each scan angle of the predetermined set of scan angles, wherein each scan line of a selected set of scan lines oriented in accordance with each scan angle passes through each of (a) a character pixel-run including at least one character pixel and (b) at least one background pixel;
• (ii) generate pixel-run representations of the character image from each angle of a predetermined set of algorithmic scan angles;
• (iii) distinguish background pixels from character pixels within an overall character image including an image boundary;
• (iv) identify and extract specified types of connected regions of (a) character pixels and (b) background pixels, the specified types of connected regions of character pixels including bars, each of which bars comprises two or more laterally adjacent character pixel-runs of at least one character pixel each, the connected regions of background pixels including lakes and bays, wherein a lake comprises laterally adjacent runs of background pixels within a connected region of background pixels that does not touch the image boundary and a bay comprises a non-boundary-touching subset of background pixel-runs within a connected region of background pixels that does touch the image boundary;
• and
• (v) generate direction-specific data concerning the extraction of bars, lakes and bays uniquely ascertainable from each of the predetermined algorithmic scan angles.

A method of recognizing an unknown object character comprising:

• capturing an image of the object character and storing the character image in a data storage device, the character image comprising character pixels within a field of contrasting background pixels, with the field of background pixels being bounded by an image edge comprised of edge pixels;
• providing an image-scanning algorithm configured to:
• (i) scan the character image along a plurality of parallel scan lines oriented each scan angle of the predetermined set of scan angles, wherein each scan line of a selected set of scan lines oriented in accordance with each scan angle passes through each of (a) a character pixel-run including at least one character pixel and (b) at least one background pixel;
• (ii) generate pixel-run representations of the character image from each angle of a predetermined set of algorithmic scan angles;
• (iii) distinguish background pixels from character pixels within an overall character image including an image boundary;
• (iv) identify and extract specified types of connected regions of (a) character pixels and (b) background pixels, the specified types of connected regions of character pixels including bars, each of which bars comprises two or more laterally adjacent character pixel-runs of at least one character pixel each, the connected regions of background pixels including lakes and bays, wherein a lake comprises laterally adjacent runs of background pixels within a connected region of background pixels that does not touch the image boundary and a bay comprises a non-boundary-touching subset of background pixel-runs within a connected region of background pixels that does touch the image boundary;
• and
• (v) generate direction-specific data concerning the extraction of bars, lakes and bays uniquely ascertainable from each of the predetermined algorithmic scan angles;
• rendering the character image accessible to the image-scanning algorithm and executing the algorithm to generate direction-specific extraction data;
• assembling a character feature vector based on the direction-specific extraction data;
• and
• communicating the assembled feature vector to character recognition apparatus configured to recognize a character corresponding to an assembled feature vector at least partially on the basis of data relating to bars, lakes and bays indicated in the feature vector.