JPH04104384A - Character recognizing device - Google Patents

Abstract

PURPOSE:To automatically correct erroneous recognition between analogous characters by specifying the position and dimension of a character to be recognized from among picture data and comparing these information with prescribed reference information. CONSTITUTION:After a character original 1 s read by a photoelectric conversion device 2, it is amplified by an amplifier 4, converted into binary digital data by a binarization circuit 5, and taken out the dimensional and position information for each character with the aid of a character segmenting part 6 to perform normalization and characteristic extraction of character. Then, a character is recognized by a dictionary 53 for identification stored in a ROM 9. Next, the decision is made and the result of the identification is corrected by a decision reference information storage part 64 of a RAM 10. The result of correction displayed on a display part 14 is corrected by hand, and the judgement of the correction is reflected on the decision reference information and utilized to the processing after that. Thus, the erroneous recognition of character picture after normalization can be automatically corrected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character recognition device, and particularly to a character recognition device that automatically corrects, for example, erroneously recognized characters.

[Prior Art] FIG. 4 is a flowchart showing character recognition processing in a conventional general character recognition device. Such a character recognition device uses a photoelectric conversion device such as an image scanner to read a document and convert it into binary image data (S41
). Subsequently, in step S42, a character image for one character is cut out from the input image converted into a binary image. This cutting is performed by first extracting a row from the pixel distribution in the row direction, and then examining the pixel distribution in the direction perpendicular to the row. Next, in order to absorb variations in the size of the cut out character images, normalization is performed in step S46. Following this,
Feature extraction of the normalized character image is performed in step S47. For the normalized image from which features have been extracted in this way, in step S48, the degree of similarity is calculated while referring to the identification dictionary 53 prepared in advance, and the character with the highest degree of similarity is selected as the recognition result. Finally, the recognition result is displayed in step S51.

[Problem to be solved by the invention] However, in the above conventional example, due to normalization, for example,
°“ °° (punctuation mark) and 0 (o), as well as °“, (comma) and °°゛°, as shown in Figure 5.

(apostrophe) and other characters (including symbols) whose character images are extremely similar after normalization are generated as a result.
There is a drawback that such similar characters may be misrecognized.

The present invention has been made in view of the above conventional example, and an object of the present invention is to provide a character recognition device that can automatically correct misrecognition of similar characters.

[Means for Solving the Problems] In order to achieve the above object, the character recognition device of the present invention has the following output. That is, a character recognition device inputs a document with characters drawn on it, reads the characters as image data, recognizes the image data as characters, and identifies the position and size of the characters from the image data. a character recognized from the image data based on a comparison result from the comparison means; A character recognition device characterized in that it has a modification means for modifying the predetermined reference information, and an updating means for updating the predetermined reference information.

[Operation] With the above output, the present invention specifies the position and size of the character recognized from the image data, and compares this information with predetermined reference information to identify and correct the character. Operate.

[Embodiments] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 is a block diagram showing the output of a character recognition device capable of automatically correcting erroneously recognized characters, which is a typical embodiment of the present invention. In FIG. 1, the character recognition device consists of an image reading section and a character detection/correction section. Furthermore, the image reading section includes a photoelectric conversion device 2 such as an image scanner.
, the scan controller 3 and the amplifier 4.

Further, the character detection/correction section includes a binarization circuit 5, a character extraction section 6, a normalization section 7, a CPU 8, a ROM 9, and a RAM 10.
1 input/output control section 11, feature extraction section 12, manual input section 13
, are output from the display section 14 and the CPU bus 15.

Here, in addition to the identification dictionary 53, the ROM 9 includes standard character size calculation, size/position information retrieval, identification calculation,
A processing program that performs various processes such as correction determination and automatic correction, which will be described later, is stored therein.

In such a character recognition device, a character document 1 is read by a photoelectric conversion device 2 such as an image scanner, the output thereof is amplified by an amplifier 4, and a binarization circuit 5 converts an analog signal into binary digital data. is converted to Next, each character is cut out by the character cutting unit 6, and after extracting character size and position information, which will be described in detail later, normalization and character feature extraction are performed. Next, ROM9
The character is recognized by selecting the character most similar to the normalized character while referring to the identification dictionary stored in the .

Next, character recognition and automatic correction processing for misrecognized characters in this embodiment will be explained using the flowchart shown in FIG. However, in the flowchart shown in Figure 2, the same processing steps as in the conventional example are given the same process numbers,
Further, the explanation will be omitted as it is based on the prior art.

First, in steps 810 to 14, image data read by an image scanner or the like, amplified by an amplifier, and binarized by the binarization circuit 5 is converted into a digitized image by the character cutting unit 6.
After each character is cut out, standard character size calculation is performed in step S18.

Calculating the standard character size means, for example, selecting a character with the maximum character height in one line of characters included in the input and binarized image data. Next, in step S20, the character height of the character is set as a reference value (H), the value of H is calculated for each line, the character height of each character image to be recognized is set to h, and h/H is calculated.
Calculate the value of . This result is stored in the size information storage section 60 of the RAM 1O. Further, in step S22, as shown in FIG. 3, the distance (1) from the top of the character image to the top of the character image is calculated using the area data of the clipping frame 70 clipped by the character clipping unit 6 and the character height of each character image. The distance (b) from the bottom of the frame to the bottom of the character image is determined, the values of t/H and b/H are calculated, and the results are stored in the position information storage section 62 of RAM10.

Next, after identifying the most similar character using the identification dictionary 53 for the normalized and feature-extracted image data, it is determined in step S30 whether the character requires modification. . Here, the modification determination includes the most similar character selected from the identification dictionary 53, the size information stored in the size information storage section 60, and the position information storage section 6.
The position information stored in 2 and the determination criterion information stored in the determination criterion information storage section 64 of RAM10 are used to perform the following determination.

For example, if the character with the highest degree of similarity selected from the identification dictionary 53 (hereinafter referred to as the first candidate) has (1) similar characters that differ only in size (e.g.
° or °゛), the threshold value U, which is the criterion information regarding the size of the first candidate character, is compared with the value of h/H stored in the size information storage unit 60, and if h/H<U, the character is lowercase. , if U≦h/H, it is determined to be an uppercase letter.

Then, in the subsequent step S32, if the first candidate is an uppercase letter and the determination result is a lowercase letter, the identification result is corrected to a lowercase letter. Thereafter, the corrected recognition result is output to the display section 14 via the input/output control section 11.

(2) If there are similar characters that differ only in their positions ((column: “”’ (apostrophe) and “’,” (comma)), t/H and b stored in the position information storage unit 62
From the value of /H, p = (t/H) - (b/H) is calculated, and the reference value P, which is the determination reference information regarding the position, is calculated.
(P>0) and Q (Q<O), if p>P, it will be placed at the bottom of the character cutting frame 70, if p<Q, it will be placed at the top of the character cutting frame 70, and if Q≦p≦P, it will be placed in the character cutting frame 70. It is determined that a character exists in the middle part of 70.

Then, in step S32, if the first candidate is a lower character and the determination result is other than the lower character, the identification result is corrected by the determination result.Then, the revised recognition result is used for input/output control. It is output to the display section 14 via the section 11.

Furthermore, after displaying the corrected recognition result on the display unit 14, in step 836, errors in the recognition result after automatic correction are manually corrected based on the correction determinations made in steps 330 to S32. While looking at the recognition results displayed on the display unit 14, the operator uses manual input from the manual input unit 13 to correct lowercase and uppercase letters, correct similar characters due to differences in position, etc. Manual corrections are made in this way. The modified results are displayed on the display section 14 again.

The manual correction of lowercase and uppercase letters and the correction of similar characters due to differences in position are reflected in the values of the criterion information, such as U, P, and Q, and are used in subsequent processing.

Therefore, according to this embodiment, not only similar characters that differ only in character size or position, such as kana and alphabetic characters, but also characters that differ in position and size, such as °", ゛ and °" °゛, can be used. Misrecognition can also be automatically corrected for characters that are different but have similar character images after normalization.

Further, in this embodiment, a case has been described in which the size and position information in the height direction of a character are used as character information for correcting misrecognition, but the present invention is not limited to this. For example, the present invention can be applied using information on character width in addition to information on character height. In this case, in the standard character size calculation of this example, the character with the maximum character width in one line is selected as the standard value (W), and it is not possible to distinguish only by the character height such as "-°゛" and "-゛". For characters that cannot be filled out, correction determination can be made based on the value of w/W for the character width (w) and the threshold value of the determination criterion information. Further, for example, the present invention can also be applied by using the ratio of character height (h) to character width (w) or the area of a character circumscribed rectangle hXw as a reference. This makes it possible to automatically correct characters due to differences in character width, such as half-width characters and full-width characters.

Further, in this embodiment, the standard character size is defined as the maximum character height in each line, but the present invention is not limited to this. For example, a character for obtaining a reference character size may be specified by external input from a manual input unit such as a mouse.

[Effects of the Invention] As explained above, according to the present invention, there is an effect that misrecognition of similar characters can be automatically corrected. Furthermore, the result of manual correction is reflected in subsequent automatic character corrections, improving the ability to automatically correct characters, which also has the advantage of reducing the manual work of correcting erroneous recognition.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing the output of a character recognition device that is a typical embodiment of the present invention, Fig. 2 is a flowchart showing character recognition and automatic correction processing of misrecognized characters, and Fig. 3 is a block diagram showing the output of a character recognition device that is a typical embodiment of the present invention. FIG. 4 is a processing flowchart of a conventional character recognition device; FIG. 5 is an explanatory diagram of normalization according to a conventional example. In the figure, 1... text manuscript, 2... photoelectric conversion device, 3...
...Scan controller, 4...Amplifier, 5...Binarization circuit, 6...Character cutting section, 7...Normalization section, 8
...CPU, 9...ROM, 10...RAM, 1
1... Input/output control unit, 12... Feature extraction unit, 13.
...Manual input section, 14...Judgment criteria information, 15...
CPU bus, 53...Identification dictionary, 60...Size information storage section, 62...Position information storage section, 64...Judgment criterion information storage section.

Claims (1)

[Scope of Claims] A character recognition device that inputs a document on which characters are drawn, reads the characters as image data, and recognizes the image data as characters, the device comprising: identifying means for identifying the position and size information outputted from the identifying means, comparing means for comparing the position and size information outputted from the identifying means with predetermined reference information, and based on the comparison result from the comparing means, from among the image data. A character recognition device comprising: a modification means for modifying a recognized character; and an updating means for updating the predetermined reference information.