JPS61279989A - System for correcting recognized result - Google Patents

Abstract

PURPOSE:To improve a correction processing speed by comparing the normalized basic feature of each character candidate with that of a character to be input ted, and outputting and displaying the candidate in the order of a higher coinci dence. CONSTITUTION:A normalized feature information part 10 stores basic features of a normalized character and a normalized graphic corresponding to a character code and a graphic one. A comparing part 11 compares the basic features stored in the information part 10, which correspond to character and graphic codes outputted from a correcting dictionary 7 with the normalized basic features of character and graphic images to be recognized, and outputs the character and graphic with the higher coincidence to a display part 5. thus, with respect to a reading input from a keying part 5, target characters with the 1st and 2nd priorities can be outputted by displaying candidate characters given to an operator, thereby improving the correction processing speed.

Description

[Detailed Description of the Invention] [Summary] In a character/figure recognition device including kanji, basic feature information for a normalized image corresponding to a character/figure code is stored, and the normalized character/figure image to be recognized is The basic features are compared with the basic features corresponding to the character/figure codes output from a correction dictionary (kana-kanji conversion dictionary, etc.) for the reading input, and the most similar character/figure code is displayed. This makes it possible to display the target character preferentially and improve the correction efficiency.

"Industrial Application Field" The present invention relates to correction of recognition results in character and figure recognition devices, and particularly relates to a method for correcting recognition errors in recognition of Chinese characters and the like.

[Prior Art] Conventional correction methods for recognition results in character and figure recognition devices have basically been configured as processing independent of recognition.

That is, when recognizing characters such as kanji, if a recognition error occurs in the recognition result, as a means to correct this, input the reading of the kanji by keystroke in order to extract the kanji code to be corrected. A group of kanji corresponding to this reading was sequentially pulled out from a pre-configured ``reading-kanji conversion dictionary'' and displayed on the display, and the operator selected one from among these to determine the corrected kanji code. .

FIG. 3 is a block diagram showing an example of the configuration of a recognition device that performs kanji recognition.

In Fig. 3, 2 is the binarization from an external device consisting of a photoelectric conversion unit that optically scans the medium on which text/graphic information exists, and a preprocessing unit that performs preprocessing on the obtained image information. This is an image memory that stores image data.

Reference numeral 3 denotes a format definition body consisting of a data group specifying positional information of characters and figures in advance, and numeral 4 denotes a recognition dictionary in which a group of characteristics of characters and figures to be processed are set.

Reference numeral 5 denotes a display section for displaying the character/figure code of the recognition result and the correction instruction character/figure code, and 6 is the keypad section for giving an instruction as a means of correcting the character/figure code of the recognition result. When editing, input in kana characters.

Reference numeral 7 denotes a correction dictionary configured in advance to output a kanji code corresponding to the kana character code string from the keypad 6, that is, a ``reading-kanji conversion dictionary''.

Reference numeral 8 denotes a search unit that searches this correction dictionary 7.

FIG. 4 is a diagram illustrating a conventional example of correction using the kana-kanji conversion method.

In FIG. 4, (a) shows the characters on the medium to be recognized, and (b) shows the recognition process.

(hl) and (b2) show the principle of the recognition process in (b).

(bl) is image data stored in the image memory and cut out one character at a time, and has a cutout size of mXn dots.

(b2) normalizes this to QXQ dots, vertically, horizontally,
and diagonal line density patterns extracted as features are shown.

Recognition processing is performed by comparing these line density features with line density features stored in a recognition dictionary.

Fig. 4(c) shows that as a result of comparison with the recognition dictionary, the following candidates were displayed in the order of highest matching: 叉, ゆう, いん, MATATA, and sentences, but the true category to be recognized was ``Jō''. It does not exist, indicating that a recognition error occurred.

(d) indicates that the operator inputs the pronunciation "JIYO" from the keypad for correction.

(e) shows the search unit searching the correction dictionary (reading-kanji conversion dictionary) based on the reading input "Jiyou".

<r> is an enlarged view of one part of the correction dictionary in (e), indicating the candidate group corresponding to the wide-reading input ``Jiyou'', such as 上, 行, sei, etc., and the target character. A certain “length” is n
It shows that it exists in the second place.

In this way, the character candidates displayed for Yomihiro input depend on the order in which they are stored in the correction dictionary, and in some cases, it may take many steps to obtain the target character. there were.

[Problems to be Solved by the Invention] As explained above, in the conventional correction method, there are generally multiple kanji codes extracted depending on the reading.
Therefore, there is a problem that many selection operations are required until the desired kanji code appears, resulting in poor operability.

Furthermore, from the operator's perspective, a kanji code different from the target kanji code is output, leading to a decline in the evaluation as a recognition device.

The present invention seeks to provide a novel correction method that eliminates these problems.

[Means for Solving the Problems] FIG. 1 shows a block diagram of the principle of a recognition result correction method according to the present invention.

In FIG. 1, the same reference numerals as in FIG. 3 indicate the same objects.

In FIG. 1, reference numeral 10 denotes a normalized feature information section, which stores basic features of normalized characters and graphics that correspond to character and graphics codes. For example, characters and figures to be recognized are normalized into QXQ dots in advance, a large amount of density information is collected for each, and the average and optimal line density information is calculated from these.
It is stored in correspondence with each character/graphic code.

Reference numeral 11 denotes a comparison unit that compares the basic features stored in the feature information unit 10 corresponding to the character/figure codes output from the correction dictionary 7 with the normalized basic features of the character/figure image to be recognized. Then, the characters/figures with the highest political change are outputted to the display unit 5 in order.

[Operation] As explained above, the present invention is capable of outputting and displaying character candidates output from a correction dictionary in response to reading input in the same order when correcting recognition results using the kana-kanji conversion method. Instead, the normalized basic features of each character candidate are compared with the normalized basic features of the input target character, and output and displayed in descending order of political change.

As a result, the target character can be output in the first or second highest order in the display of candidate characters given to the operator in response to a wide reading input, and the correction processing speed can be improved.

In other words, as a means of indexing a character/figure code that is more similar to the original character/figure image from a correction dictionary,
□ Store the normalized feature information of figures in advance and compare this with the basic features during recognition of characters and figures to be recognized, and check the original purpose and configuration of the recognition dictionary and correction dictionary. This minimizes the reduction in correction efficiency due to differences.

[Example] FIG. 2 is a conceptual block diagram of an embodiment of the present invention.

In FIG. 2, (a) shows the feature information section, and each character
Corresponding to the figure code, a table (or line density code) of the line density pattern in the horizontal (X) direction and vertical (Y) direction of each character is stored as a basic feature of normalized characters and figures. It is a 48-byte table.

(b) shows the horizontal (X) direction and vertical (
The line density pattern is 48 bytes in the Y) direction.

(c) shows that the correction dictionary is searched by inputting the pronunciation "Jiyou" from the keypad, and the character codes of 上, jo, and sei are output as i+i+1 and i+2, respectively.

In (d), the comparison unit calculates the distance between the recognition target character output from (b) and the linear density pattern for the i-th character code output from (c), and if they are approximate, this is the distance. is output, and if they are not approximate, the process moves to the i+1th character code and is compared again. In this way, the approximate "length" is output.

[Effects of the Invention] As described above, according to the present invention, there is provided a kana-kanji conversion method in a character/figure recognition device including kanji.

Significantly improves correction efficiency when correcting recognition results using
.

It can be expected that this will improve the performance, and its practical effects will be significant.

[Brief explanation of the drawing]

Fig. 1 is a principle block diagram of the present invention, Fig. 2 is a conceptual block diagram of an embodiment of the present invention, Fig. 3 is a block diagram of a conventional example, and Fig. 4 is a diagram explaining a modification method of the conventional example. . In the drawings, 1 is a recognition control unit, 2 is an image memory, 3 is a format 1 definition body, 4 is a recognition dictionary, 5 is a display unit,
6 is a keypad, 7 is a correction dictionary,
8 is a search section, 9 is an output section, 10 is a feature information section, and 11 is a comparison section.

Claims (1)

[Scope of Claims] A format consisting of an image memory (2) that stores image information obtained by scanning a medium containing character/graphic information by optical means, and a data group in which positional information of characters/graphical information, etc. are specified in advance. A definition body (3), a recognition dictionary (4) that stores the feature group of characters and figures to be processed, and a display unit (5) that outputs and displays the character and figure codes of the recognition results.
), a keypad (6) for giving an instruction as a means of correcting the character/figure code of the recognition result, and a keypad (6) configured in advance to output a character/figure code corresponding to the input code string from the keypad. A correction dictionary (7), a search unit (8) that searches the correction dictionary (7), an output unit (9) that outputs character/graphic codes resulting from recognition and correction onto an output medium, and each of the above units. A recognition control unit (1) that controls interfaces with external devices and a feature information unit (10) that stores basic characteristics of normalized characters and figures corresponding to character and figure codes. ) and the correction dictionary (
7) The basic features stored in the feature information section (10) corresponding to the character/graphic code output from
Comparison unit (1) that compares the normalized basic features of the graphic image
1), and the result of comparison by the comparison unit (11),
A recognition result correction method characterized in that the most similar character/figure code is displayed.