Optical character reading method and system for a document with ruled lines and its application

A method of recognizing characters of a document in a system having means for reading the document and storing read data as an input image of pels, the method comprising:

• the step of obtaining said characters from said document;
• the step of extracting bounding boxes which contact respective characters in every side and character patterns defined by respective bounding boxes;
• and
• the steps, executed for each of said characters, of:
• from said character pattern, extracting contour information and ground information as fundamental features;
• extracting at least one kind of statistical features on the basis of said fundamental features;
• extracting at least one kind of structural features on the basis of said fundamental features and said character pattern;
• on the basis of said extracted statistical features, finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree of similarity and degree(s) of similarity associated with said at least one provided candidate;
• providing a final candidate for said character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and said bounding box;
• wherein said document includes ruled lines;
• wherein said step of obtaining said characters from said document includes the step of separating said characters from said ruled lines;
• and
• wherein said system is provided with a table comprising records associated one by one with a plurality of entry forms each having a column that accepts a set of predetermined character strings, each record containing features of said ruled lines, information on said column, and wherein the method further comprises the steps of:
• using one of said entry forms for said document;
• extracting ruled line features from said input image;
• retrieving a record associated with said used entry form through comparison between said ruled line features and said features of ruled lines;
• and
• applying said step of extracting bounding boxes to said column of said retrieved record.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for providing a final candidate comprises:
• means for obtaining a set of criterions concerning relevant one(s) of said extracted structural features, said set of criterions being associated with at least one of said provided candidate(s) and said provided degree(s) of similarity from said means for finding some number of candidates;
• and
• means, based on said set of criterions, for examining said relevant structural features to select said final candidate;
• means for storing confusing provided patterns and associated set of criterions, a provided pattern being a whole of said provided candidate(s) and said provided degree(s) of similarity from said means for finding some number of candidates, and wherein means for obtaining a set of criterions comprises means for obtaining said set of criterions from said storing means through a predetermined access path;
• and
• first means which is a version of said storing means which version is dedicated to printed characters;
• second means which is a version of said storing means which version is dedicated to handwritten characters;
• means for judging, from sizes of said bounding boxes, whether said characters are printed or handwritten;
• and
• means, disposed in said predetermined access path, for connecting said means for obtaining from said storing means to one of said first and second means according to said judging means.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for providing a final candidate comprises:
• means for obtaining a set of criterions concerning relevant one(s) of said extracted structural features, said set of criterions being associated with at least one of said provided candidate(s) and said provided degree(s) of similarity from said means for finding some number of candidates;
• and
• means, based on said set of criterions, for examining said relevant structural features to select said final candidate;
• wherein said means for examining said relevant structural features to select said final candidate comprises:
• means for correcting said provided degree(s) of similarity into corrected degree(s) of similarity on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and said bounding box;
• and
• means for selecting a candidate that has a largest corrected degree of similarity as said final candidate;
• and
• wherein means for correcting said provided degree(s) of similarity comprises:
• means, responsive to an examination of each of said relevant structural features, for calculating correction amounts for by incrementing or decrementing zero-preset correction amount register(s) of character(s) associated with respective relevant structural features;
• and
• means for adding, to each of said provided degree(s) of similarity, said correction amount of a candidate associated with said each of said provided degree(s).

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• and
• means for determining whether said character has been printed or handwritten, on the basis of a size of said bounding box, wherein said means for providing a final candidate includes means for changing a way of using said structural features.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• means for providing a reject signal if it is determined that said provided candidates are not reliable from a ratio between a largest degree of similarity and a second largest degree of similarity and a complex degree which indicates a complexity of said character pattern and calculated from ground information, said ground information being one of said structural features;
• and
• means for providing said provided candidates as they are if said reject signal is provided;
• wherein said means for providing a reject signal comprises means for providing a reject signal if said largest degree of similarity is smaller than a first predetermined threshold, if a ratio of said second largest degree of similarity to said largest degree of similarity is larger than a second predetermined threshold, or if said complex degree is larger then a third predetermined threshold.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding decrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said document includes ruled lines;
• wherein said step of obtaining said characters from said document includes the step of separating said characters from said ruled lines;
• and
• wherein said system is provided with a table comprising records associated one by one with a plurality of entry forms each having a column that accepts a set of predetermined character strings, each record containing features of said ruled lines, information on said column and set of predetermined character strings, wherein one of said entry forms is used for said document, and wherein the system further comprises:
• means for extracting ruled line features from said input image;
• means for retrieving a record associated with said used entry form through comparison between said ruled line features and said features of ruled lines;
• and
• means for applying said means for extracting bounding boxes to said column of said retrieved record.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• and
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for extracting contour information and ground information as fundamental features comprises:
• means for extracting contour pels constituting a contour of said extracted character pattern;
• contour direction finding means for finding one of predetermined contour directions for each contour pel from a state of neighboring pels of said each contour pel;
• margin finding means for calculating a distance, along each pel line, from each side of said bounding box to a first pel of said character pattern;
• and
• ground path finding means for calculating, along each pel line from each side to an opposing side of said bounding box, coordinates of a start pel and an end pel of each of ground paths and a number of said ground paths, each of said ground paths comprising a consecutive ground (i.e., white) pel string.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for extracting at least one kind of statistical features includes:
• means for dividing said extracted character pattern into upper and lower two areas;
• means for dividing said extracted character pattern into a left area and a right area;
• wherein said means for extracting at least one kind of statistical features further includes:
• means for counting, for each of four blocks divided by said two dividing means, contour pels of an identical contour direction category to provide a pel-count-by-contour-direction for each block, said counting being performed for each of predetermined contour direction categories;
• and
• means for dividing said pel-count-by-contour-direction by a size of said each block to provide a normalized pel-count-by-contour-direction.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and around information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for extracting at least one kind of statistical features includes:
• means for dividing said extracted character pattern into upper and lower two areas;
• means for dividing said extracted character pattern into a left area and a right area;
• wherein said means for extracting at least one kind of statistical features further includes:
• means for counting, for each of said upper and said lower areas, contour pels which first appear when horizontal pel lines scanned from each of a left and a right side to an opposing side and have an identical contour direction category to provide an outer-pel-count-by-contour-direction for each scanning direction for each of said upper and said lower areas and for counting, for each of said left and said right areas, contour pels which first appear when vertical pel lines scanned from each of a top and a bottom side to an opposing side and have an identical contour direction category to provide an outer-pel-count-by-contour-direction for each scanning direction for each of said left and said right areas;
• and
• means for dividing said outer-pel-count-by-contour-direction by a size of said each area to provide a normalized outer-pel-count-by-contour-direction.

A system for recognizing characters of a document, the system comprising:

• means for reading said document and storing read data as an input image of pels;
• means for obtaining said characters from said document;
• means, operative for each character, for extracting a bounding box and a corresponding character pattern defined by said bounding box,
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• structural feature extracting means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• wherein said means for extracting at least one kind of statistical features includes:
• means for dividing said extracted character pattern into upper and lower two areas;
• means for dividing said extracted character pattern into a left area and a right area;
• wherein said means for extracting at least one kind of statistical features further includes:
• means calculating a sum of left margins for each of said upper area and said lower area to provide upper and lower left margin sums;
• calculating a sum of right margins for each of said upper area and said lower area to provide upper and lower right margin sums;
• calculating a sum of top margins for each of said left area and said right area to provide left and right top margin sums;
• and calculating a sum of bottom margins for each of said left area and said right area to provide left and right bottom margin sums;
• and
• means for dividing each of said margin sums by a size of said each area to provide a normalized margin sum.

A system for automatically sorting a plurality of entry forms each including ruled lines and a column that accepts a set of predetermined character strings, the system comprising:

• a table comprising records associated one by one with said plurality of entry forms, each record containing features of said ruled lines, data that defines said column and said set of predetermined character strings;
• means for reading each of said entry forms and storing read data as an input image of pels;
• means for extracting ruled line features from said input image;
• means for retrieving a record associated with said used entry form through comparison between said ruled line features and said features of ruled lines;
• means for removing said ruled lines from said input image;
• means, operative for each character of a character string in an area specified by said data that defines said column of said retrieved record, for extracting a bounding box and a corresponding character pattern defined by said bounding box;
• means for extracting contour information and ground information as fundamental features from said character pattern;
• means for extracting at least one kind of statistical features on the basis of said fundamental features;
• means for extracting at least one kind of structural features on the basis of said fundamental features and each character pattern;
• means, based on said extracted statistical features, for finding some number of candidates for said character and corresponding degrees of similarity to provide at least such one of said candidates as has largest degree(s) of similarity and degree(s) of similarity associated with said at least one provided candidate;
• means for providing a final candidate for each character pattern on the basis of said at least one provided candidate, said provided degree(s) of similarity, said structural features and each bounding box;
• means for searching said set of predetermined character strings in said retrieved record for a recognized character string comprising a collection of said final candidates;
• means, responsive to a success in said searching said set, for outputting said recognized character string;
• means, responsive to a failure in said searching said set, for outputting a most similar one of said predetermined character strings as said recognized character string;
• a conversion table for associating each of said predetermined character strings with a corresponding code;
• means for obtaining a code associated with said recognized character string;
• means for printing said obtained code at a predetermined position on each of said entry forms;
• and
• means for sorting said entry forms by said printed obtained codes.