System and method for automatically classifying heterogeneous business forms

An image processing system for automatically recognizing and classifying heterogeneous business forms, comprising:

• means for storing a plurality of template features of one or more business forms;
• means for scanning one or more actual business forms;
• means for extracting features of said actual business forms, wherein said means for extracting includes means for scanning said actual business forms on a line by line basis to form black and white pixel data representative thereof, means for identifying neighboring ones of said black pixels to identify connected components having X and Y components, means for determining if said connected components form elongated line segments, means for determining whether two or more contiguous segments are part of the same line, and means for merging said segments considered to be part of the same line;
• and
• matching means for sequentially matching said extracted features with the plurality of stored template features, wherein said means for matching includes means for recognizing vertical and horizontal lines, means for adding the length of al lines found in the scanned image to form a key value, means to condense line information from said template and actual forms into single scalar values such as line and intersection counts or line length sums so as to discard templates from a candidate list, means to determine the spatial similarity between two or more line patterns, means for measuring the distribution of lines in said scanned image to condense the projection information into a binary value representative of the presence of lines in predefined region of said image, means for comparing the line length and the distance between lines, means for analyzing the line patterns in said actual and template forms and means for determining the type of actual business form.

A business form recognition system comprising

• means for storing template features of one or more business forms,
• means for scanning one or more actual business forms to form scanned data representative thereof, including means for scanning said actual business forms on a line by line basis to form black and white pixel data representative thereof and means for identifying neighboring ones of said black pixels to identify connected components having X and Y components,
• pattern recognition means for recognizing said scanned data, wherein said pattern recognition means include means for recognizing only vertical and horizontal line patterns on said scanned business forms,
• matching means for matching the recognized data with a plurality of stored forms to provide a determination of the type of the actual business form, said matching means including
• indexed dictionary access means for adding the length of all lines found in the scanned image to form a key value,
• aggregate discrimination means to condense line information from said template and actual forms into single scalar values such as line and intersection counts or line length sums so as to discard templates from a candidate list,
• intersection discrimination means to determine the spatial similarity between two or more line patterns, approximation discrimination means for measuring the distribution of lines in said scanned image to condense the projection information into a binary value representative of the presence of lines in predefined regions of said image,
• projection discrimination means for comparing the line length and the distance between lines,
• window matching means for analyzing the line patterns in said actual and template forms, and
• decision function means for determining the type of actual business form.