Character recognition apparatus and character recognition method

A character recognition apparatus comprising:

• extraction unit extracting a dotted line candidate from image data of a form;
• feature collection unit collecting a thickness, length, interval of a dotted line of a plurality of extracted dotted line candidates as feature information;
• estimation unit estimating a dotted line candidate spot in a character box being a recognition target based on the feature information of a dotted line candidate in a same character box being the recognition target or in another character box;
• elimination unit eliminating the dotted line candidate extracted by the extraction unit and the dotted line candidate estimated by the estimation unit from the image data;
• and
• character recognition unit performing recognition of a character in image data from which the dotted line candidate has been eliminated by the elimination unit,
• wherein the estimation unit estimates whether or not there is a dotted line candidate not extracted yet in the character box being the recognition target, using an interval between a plurality of dotted line candidates in the character box being the recognition target or another relevant character box, and if it is estimated that there is a dotted line candidate, image data of a position on which a dotted line is estimated to be present in the image data and image data of a reference dotted line obtained from a plurality of dotted line candidates are compared, and when an agreement rate of the image data is equal to or larger than a reference value, it is estimated that a same dotted line candidate as the reference dotted line exists on the estimated position.

A character recognition apparatus comprising:

• extraction unit extracting a dotted line candidate from image data of a form;
• feature collection unit collecting a thickness, length, interval of a dotted line of a plurality of extracted dotted line candidates as feature information;
• estimation unit estimating a dotted line candidate spot in a character box being a recognition target based on the feature information of a dotted line candidate in a same character box being the recognition target or in another character box;
• elimination unit eliminating the dotted line candidate extracted by the extraction unit and the dotted line candidate estimated by the estimation unit from the image data;
• and
• character recognition unit performing recognition of a character in image data from which the dotted line candidate has been eliminated by the elimination unit,
• wherein the estimation unit creates an interval table representing intervals between dotted line candidates using intervals between dotted line candidates according to intervals between a plurality of dotted line candidates extracted in a same character box;
• assigns a different number of votes respectively to values that are 1 time, (1/n) time an interval in the interval table;
• obtains a cumulative total of a number of votes for each interval in the interval table and estimates an interval whose cumulative total value is largest as an interval between the dotted line candidates.

A character recognition method comprising:

• extracting a dotted line candidate from image data of a form;
• collecting and storing, in storing means as feature information, a thickness, length, interval of a dotted line of a plurality of extracted dotted line candidates;
• estimating a dotted line candidate spot in a character box being a recognition target, based on the feature information of a dotted line candidate in a same character box being the recognition target or in another character box;
• eliminating the extracted dotted line candidate and the estimated dotted line candidate from the image data;
• and
• performing recognition of a character in image data from which the dotted line candidate has been eliminated,
• wherein a dotted line candidate spot and a type of a dotted line candidate in the character box being the recognition target are estimated, using the feature information of a dotted line candidate in the same character box being the recognition target or in another character box and information indicating an interval between a dotted line candidate in the character box being the recognition target and a dotted line candidate in a same character box as the character box being the recognition target or in another character box.

A character recognition method comprising:

• extracting a dotted line candidate from image data of a form;
• collecting and storing, in storing means as feature information, a thickness, length, interval of a dotted line of a plurality of extracted dotted line candidates;
• estimating a dotted line candidate spot in a character box being a recognition target, based on the feature information of a dotted line candidate in a same character box being the recognition target or in another character box;
• eliminating the extracted dotted line candidate and the estimated dotted line candidate from the image data;
• and
• performing recognition of a character in image data from which the dotted line candidate has been eliminated,
• wherein whether or not there is a dotted line candidate not extracted yet in the character box being the recognition target is estimated, using an interval between a plurality of dotted line candidates in the character box being the recognition target or another relevant character box, and if it is estimated that there is a dotted line candidate, image data of a position on which a dotted line is estimated to be present in the image data and image data of a reference dotted line obtained from a plurality of dotted line candidates are compared, and when an agreement rate of the image data is equal to or larger than a reference value, it is estimated that a same dotted line candidate as the reference dotted line exists on the estimated position.

A character recognition method comprising:

• extracting a dotted line candidate from image data of a form;
• collecting and storing, in storing means as feature information, a thickness, length, interval of a dotted line of a plurality of extracted dotted line candidates;
• estimating a dotted line candidate spot in a character box being a recognition target, based on the feature information of a dotted line candidate in a same character box being the recognition target or in another character box;
• eliminating the extracted dotted line candidate and the estimated dotted line candidate from the image data;
• and
• performing recognition of a character in image data from which the dotted line candidate has been eliminated,
• wherein an interval table representing intervals between dotted line candidates is created, using intervals between dotted line candidates according to intervals between a plurality of dotted line candidates extracted in a same character box;
• a different number of votes is assigned respectively to values that are 1 time, (1/n) time an interval in the interval table;
• a cumulative total of a number of votes for each interval in the interval table is obtained and an interval whose cumulative total value is largest is estimated as an interval between the dotted line candidates.