JPH11203411A - Document reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document reader capable of securing the stable quality by quantitatively evaluating the quality of a read document. SOLUTION: A recognition result evaluation part 6 evaluates the quality of document text data generated by an image input part 2 and a character recognition part 4 based on the recognition accuracy of respective characters. Whether the evaluated quality satisfies a prescribed qualified condition or not is judged, and when the condition is not satisfied, a parameter changing part 8 changes an image reading parameter of the image input part 2 to reread the image. Basically the feedback loop is repeated until prescribed quality can be obtained. For example, the reading resolution of the image input part 2 is used as the parameter, which is stepwisely increased according to the number of loops.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document reading apparatus which captures a document image and performs character recognition to generate document text data, and more particularly to ensuring the quality of text data obtained as a result of character recognition.

[0002]

2. Description of the Related Art For example, a document printed or handwritten on paper or the like is read from a scanner as document image data, and the document image data is processed by character recognition software operating on a general-purpose computer such as a personal computer. A document reading system that recognizes each character written in the document and generates document text data,
In recent years it is no longer unusual.

[0003] However, in Japanese documents in particular, in addition to Japanese characters such as kanji, English characters and symbols are used, and there are many types of characters used. In addition, there are many objects having a complicated character image and images having image images similar to each other. Therefore, research has been conducted on how to accurately perform character recognition.

For example, the "recognition result evaluation method" described in Japanese Patent Application Laid-Open No. Hei 6-187487 is used to evaluate the accuracy of character recognition in a character recognition system that obtains a plurality of candidate characters as recognition results. The number of strokes or the number of constituent line primes of a plurality of candidate characters is used. Since the number of strokes and the number of constituent line elements are numerical values determined for each character, the recognition result is hardly affected by differences in the character recognition method, document image quality, font, and the like. According to the method, the accuracy of character recognition is improved and stabilized.

[0005] As described in "Character Recognition Apparatus" described in Japanese Patent Application Laid-Open No. 5-189620, knowledge processing is used to find misrecognized characters. In the knowledge processing, a character string generated by a character recognition candidate as a recognition result is compared with a dictionary. For a character having a plurality of character recognition candidates, the characters are replaced and collation is performed. If the collation is successful, a process of increasing the priority of the character recognition candidate is performed.

As described above, in the prior art, character-by-character evaluation is performed to obtain a character recognition accuracy that is a certainty that a character candidate obtained by recognition is correct.
In addition, such techniques have been improved, and priorities determined thereby have been adjusted to reduce erroneous recognition.

[0007]

However, in the operation of reading a document from a scanner and performing character recognition processing on the image file to obtain document text data,
The conventional evaluation method in units of characters has a problem that the quality of document text data cannot be quantitatively evaluated.

For this reason, conventionally, generally, the user himself has
The recognized document text is visually inspected on the screen or in a printed output, and its quality is evaluated.Until satisfactory quality is obtained, the document is read again by increasing the resolution of the scanner, and character recognition processing is performed. It was necessary to repeat the work of doing. Such a document reading operation is a troublesome operation for the user, and has a problem that the efficiency is very low.

Further, in the method of acquiring document image data by increasing the resolution of the scanner from the beginning, it takes a long time to read the document, and when storing an image file due to an increase in the size of the image file, a large size is required. Problems arise, such as the need for space, and these problems are particularly acute when reading large volumes of documents.

[0010] Further, in the visual evaluation by the user, since the evaluation criterion varies from user to user and from case to case,
There is a problem that it is difficult to perform a stable objective evaluation of the document reading result, and the reliability of the obtained document text data may not be stable.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and objectively evaluates the quality of generated document text data, thereby ensuring reliability without imposing a burden on a user. An object of the present invention is to provide a document reading apparatus capable of quickly obtaining document text data and improving document reading work efficiency.

[0012]

According to a first aspect of the present invention, there is provided a document reading apparatus which evaluates the reading quality of document text data based on the character recognition accuracy evaluated and output for each character by a character recognition unit. A recognition result evaluating unit that determines whether an evaluation result satisfies a predetermined eligibility condition; and a parameter changing unit that changes an image reading parameter in an image input unit when the eligibility condition is not satisfied. .

According to the present invention, the recognition result evaluating unit evaluates the quality of the generated text data on a document basis, which is a larger unit, using the character recognition accuracy on a character basis. Here, the unit of the document may be a single form, or may be a logical group of a plurality of forms. The parameter changing unit changes the image reading parameters of the image input unit based on the reading quality evaluation result output by the recognition result evaluating unit. That is, the evaluation result of the reading quality is fed back to the image input unit, and thereby the feedback control for improving the reading quality to satisfy the qualifying condition is performed.

In the document reading apparatus according to a second aspect of the present invention, the image input unit generates document image data for the same document again based on the changed image reading parameters, and performs the character recognition. Output to the unit.

According to the present invention, a series of steps of changing image reading parameters, generating document image data, and generating document text data by character recognition processing are performed on the same document until a predetermined eligibility condition is satisfied. The process is repeated. Thereby, the quality of the document text data of a predetermined level or more for each document is stably ensured. Here, the repetition of the processing may be performed for each single form, or may be performed in units of a logical group of a plurality of forms.

In a preferred aspect of the present invention, the image reading parameter includes a resolution for reading the document image. In another preferred aspect, the image reading parameter includes a density threshold in reading the document image.

According to a third aspect of the present invention, in the above-mentioned invention, the image reading parameter includes a resolution for reading the document image, and the parameter changing unit increases the resolution.

According to the present invention, the image input unit has a plurality of types of reading resolutions, and when a certain resolution does not satisfy the qualifying condition, the image is read at a higher resolution. Generally, by increasing the resolution, the recognition accuracy of each character is improved, and the reading quality of each document is also improved. Therefore, for example, the resolution that achieves the read quality that satisfies the qualifying condition may be retained and used continuously for reading the subsequent documents until it is newly determined that the qualifying condition is not satisfied.

[0019] In a further preferred aspect of the present invention, there is provided a parameter initializing section for setting a predetermined initial value as the image reading parameter in the first reading of the document, wherein the parameter initializing section sets the resolution of the resolution. A predetermined coarse resolution is set as an initial value. In this aspect, the resolution is reset to the coarse resolution for each document.
The coarse resolution is a lower value of any of the multiple resolutions of the image input unit, and can be set to, for example, the lowest resolution. Generally, when the image input unit is operated at a high resolution, it takes time to generate document image data. On the other hand, for each document, the resolution is set to a low value as the initial value,
By increasing the resolution as needed, that is, only when the eligibility condition is not satisfied, the throughput of the document reading process can be improved.

In the document reading apparatus according to a fourth aspect of the present invention, the recognition result evaluating section calculates a character recognition accuracy average value obtained by averaging the character recognition accuracy for each character included in the entire region or a predetermined partial region of the document. Then, the eligibility of the evaluation result is determined based on a comparison between the average value of the character recognition accuracy and a predetermined reference value.

In the document reading apparatus according to a fifth aspect of the present invention, the recognition result evaluating unit may determine a ratio of a character having the character recognition accuracy in a predetermined range among characters included in the entire area or the predetermined partial area of the document. Then, the eligibility of the evaluation result is determined based on the comparison between the ratio and a predetermined reference ratio.

According to the present invention, in a predetermined target area of a document, a ratio between the number of characters whose character recognition accuracy is a value within a predetermined range and the total number of characters included in the area is obtained. This ratio is compared with a predetermined reference ratio to determine whether the read quality evaluation result is qualified or not. The predetermined range of the character recognition accuracy can be specified by a set of upper and lower limits or specified by a threshold. For example, when the ratio of characters having a character recognition accuracy equal to or higher than a predetermined threshold is equal to or higher than a predetermined reference ratio, it is determined that the evaluation result is qualified.

[0023]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic block diagram of a document reading apparatus according to an embodiment of the present invention. This apparatus includes an image input unit 2, a character recognition unit 4, a recognition result evaluation unit 6, a parameter change unit 8, an image input control unit 10, a quality condition input unit 1,
2. It is configured to include the recognition result output unit 14, reads the document 20, and generates document text data.

The image input unit 2 reads an image of the document 20 and generates document image data. As the image input unit 2, for example, an optical scanner is used. For example, the image input unit 2 may be provided with an automatic paper feed mechanism, and may be configured to automatically and sequentially read a form as the document 20.

The character recognizing section 4 recognizes a character image included in the document image data generated by the image input section 2 and converts the image into a corresponding character code.
Generate document text data.

FIG. 2 is a schematic diagram showing an example of a character recognition process performed by the character recognition section 4. In the character recognition unit 4 of this example, a plurality of character candidates are obtained for each character image, and each character candidate is given character recognition accuracy as an index indicating the probability of recognition. The character recognition accuracy indicates the probability that the character candidate is a character described in the document, and does not mean that a character candidate with a higher accuracy is always a correct character than a character with a lower accuracy. If a plurality of character candidates are obtained, they are prioritized. This priority can be simply determined according to the degree of the character recognition accuracy, and can be adjusted based on knowledge processing as in the above-described prior art.

In the example shown in the figure, the image 30 is generated by the image input unit 2. The image 30 shows an example in which, for example, an image of a character string that is a “vector” on the actual document 20 is captured with noise included in the second character “ku” of the character string due to dust on the document, for example. ing. When the document image data is input to the character recognizing unit 4, the character recognizing unit 4 recognizes the image data by a character recognizing process as shown as a recognition result 40. The recognition result 40 is, for example, "be" (Katakana), "Be" (Hiragana), "Pe" (Katakana), as the first to fifth character candidates for the first character of the character string.
Recognition results of “ぺ” (Hiragana) and “Yo” were obtained, indicating that a value of “0.788184” was obtained as the character recognition accuracy of the first candidate. In addition, the recognition result 40 indicates that for the second character of the character string, “ta” (Katakana), “C” (Katakana), and “G”
(Katakana), "evening" (Kanji), and "Da" (Katakana) are obtained, and the character recognition accuracy of the first candidate is "0.
698382 ". That is, the second character is recognized as an incorrect character" ta "instead of the correct character" ku "due to noise. A character image generated with noise (in the above example, an image in which noise is added to “ク”) does not have the perfect shape of the original correct character, and generally, a character that is incorrectly recognized (in the above example, Neither does it have the perfect form of "ta"). Therefore, generally, the recognition accuracy of a character erroneously recognized as the first candidate is low.

The recognition result evaluation unit 6 receives from the character recognition unit 4 a recognition result including the character recognition accuracy having such a property. The recognition result evaluation unit 6 evaluates the quality of the entire recognition result of the document 20 based on the character recognition accuracy obtained from the character recognition unit 4.

A condition (qualifying condition) serving as a criterion for judging that the quality is qualified is set from the quality condition input unit 12 to the recognition result evaluating unit 6. For example, the user inputs the eligibility condition to the quality condition input unit 12 and the quality condition input unit 1
2 can set this to the character recognition unit 4.

When an evaluation result satisfying the eligibility condition is obtained, the recognition result evaluation unit 6 outputs the obtained document text data to the recognition result output unit 14. On the other hand, when the evaluation result satisfying the desired eligibility condition is not obtained, the recognition result evaluating unit 6 notifies the parameter changing unit 8 of the result.

The parameter changing section 8 changes parameters when the image input section 2 reads a document image. In this configuration, the image reading parameters are directly set in the image input unit 2 by the image input control unit 10. In this configuration, the parameter changing unit 8 is configured to generate and pass a new parameter value to the image input control unit 10 or to generate and pass a difference amount from the current value. Can be done. The image reading parameters are, for example, the reading resolution of the document image in the image input unit 2 and other density thresholds used as criteria for determining whether each point on the document is determined to be white or black.

The image input control unit 10 controls the operation of the image input unit 2. For example, as described above, when image reading parameters are set in the image input unit 2, start / stop of a reading operation is controlled, and when the image input unit 2 has an automatic paper feed mechanism, , Are also controlled.

The recognition result output unit 14 has a function of displaying and recording the generated document text data. For example, the recognition result output unit 14 may be configured using a display for displaying a screen, a storage device such as a magnetic disk device capable of storing electronic document text data in a file format, a printer for outputting to paper or the like, or the like. Can be.

As described above, in the present apparatus, the image input unit 2, the character recognition unit 4, the recognition result evaluation unit 6, the parameter change unit 8, and the image input control unit 10 form a feedback loop.

The document 20 is, for example, a form, that is, a set of sheet-shaped documents. In particular, when the image input unit 2 has an automatic paper feed mechanism, the plurality of forms are set in the apparatus as a stack. Would be common.

The character recognition unit 4, the recognition result evaluation unit 6, the parameter change unit 8, the image input control unit 10, and the quality condition input unit 12 are, for example, a central processing unit (CPU: Central Proc).
essing unit). That is, an example of the present apparatus can be configured by connecting the scanner apparatus to a general-purpose computer such as a personal computer.

FIG. 3 is a processing flowchart for explaining the processing in the present apparatus. Here, an example is given in which the eligibility condition is set as “the number of characters whose character recognition accuracy of the first candidate is 0.9 or more occupies 90% or more of the recognized document text.” Will be described. It is assumed that only the reading resolution among the image reading parameters is considered. Further, it is assumed that the image input unit 2 has an automatic paper feeding mechanism and can sequentially read the sheets set in a stack one by one.

First, the user sets eligibility conditions in the quality condition input section 12. The setting of this eligibility condition is, for example,
A threshold “0.9” of the character recognition accuracy, and a reference ratio “9” which is a threshold relating to a ratio of characters whose character recognition accuracy exceeds the threshold.
0% ". The input eligibility condition is transferred to the recognition result evaluation unit 6 and stored (S5).
0). Thereafter, the document reading operation of the apparatus is started.

The image input control unit 10 controls the automatic sheet feeding mechanism, and every time a form is newly set in the image input unit 2,
The initial values of the image reading parameters are set in the image input unit 2 (S55). This initial value is stored in the image input control unit 10 in advance. For example, the image input control unit 10 sends a 20
Set 0 dpi. Then, the image input control unit 10 instructs the image input unit 2 to read the form set from the automatic sheet feeding mechanism.

The image input section 2 optically reads the image of the form at the set resolution and outputs document image data (S60). This document image data is passed to the character recognition unit 4 as an image data file. The process of combining the document image data into a file may be performed by the image input unit 2 itself. However, when the image input unit 2 is a scanner connected to a computer, a processing unit (not shown) on the computer is generally used. The input data from the scanner will be filed and passed to the character recognition unit 4.

The character recognition section 4 performs a character recognition process on the input image file (S65). That is, the character recognition unit 4 detects a data portion representing a character image included in the image file, and performs character recognition on each character image. As a result, when a plurality of character candidates are detected for one character image, the character recognition unit 4 determines a priority order among the candidates and obtains at least the character recognition accuracy of the first candidate.

The recognition result of each character image by the character recognition unit 4 is input to the recognition result evaluation unit 6. The recognition result evaluation unit 6 first evaluates the recognition result (S70). In this evaluation, in the example described here, the magnitude of the character recognition accuracy of each character and the threshold “0.9” set by the quality condition input unit 12 are determined, and the number of characters exceeding the threshold and the number of characters in the form are determined. The ratio to the total number of characters is calculated. This ratio is compared with a reference ratio “90%” set in the quality condition input unit 12.

If the detected ratio is equal to or higher than the reference ratio (S
75) The character recognition quality of the read form is determined to be qualified, and the document text data input from the character recognition unit 4 to the recognition result evaluation unit 6 is output to the recognition result output unit 14 (S80). . The recognition result output unit 14 displays input document text data according to the function,
Memorize and print.

On the other hand, if the detected ratio is lower than the reference ratio, it is determined that the quality is inappropriate (S75).
In this case, the recognition result evaluator 6 includes the parameter changer 8
Is instructed to change the reading resolution.
Generally, erroneous character recognition occurs when the reading resolution is insufficient. Therefore, upon receiving the instruction, the parameter changing unit 8 adds a preset change width to the current resolution value, and outputs the increased resolution value to the image input control unit 10 (S85). The change width is, for example, 100
It can be set to dpi. In this case, if it is determined that the first reading is inappropriate, the initial value of the resolution 200
300 dpi obtained by adding the change width to the dpi is output to the image input control unit 10.

When the changed resolution is input from the parameter changing unit 8, the image input control unit 10 sets the changed resolution in the image input unit 2, and re-reads the form currently set in the image input unit 2 again. Instruct to read. Image input unit 2
Outputs the document image data read at the new resolution to the character recognition unit 4 as in the previous case. Thereafter, the above-described processing is repeated. That is, if the recognition result evaluation unit 6 determines that the quality is inappropriate again, the parameter changing unit 8 further increases the resolution by 100 dpi, and increases the resolution by 400 dpi.
Is read.

That is, in the present apparatus, the recognition result evaluation unit 6 determines that the quality is appropriate (S75), or until the resolution reaches the upper limit of the performance of the image input unit 2 (S90).
The document reading is repeated with the resolution incremented.

If the recognition result evaluation section 6 determines that the quality is appropriate, the document text data is output to the recognition result output section 14 and simultaneously to the image input control section 10 as described above. Is also notified. Image input control unit 1
When 0 receives this notification, it controls the automatic paper feed mechanism to set the next form, resets the resolution to the initial value, and reads a new form.

Also, when the resolution value passed from the parameter changing unit 8 exceeds the upper limit of the performance of the image input unit 2, the image input control unit 10 similarly resets the resolution and creates a new form. Execute the reading process. For example, in this case, the image input control unit 10 notifies the recognition result evaluation unit 6 that the passed resolution exceeds the upper limit, and the recognition result evaluation unit 6 performs the most recently obtained document based on the notification. The text data is output to the recognition result output unit 14. Although the document text data output here is not data whose quality has been determined to be qualified, it is the data of the best quality that can be obtained by this apparatus, and therefore this apparatus outputs this data. In this case, it may be desirable to output data indicating that the quality is ineligible and the quality.

In the above description, the initial value of the reading resolution and the change width of the reading resolution per change in the parameter change are described in the above description, respectively.
Although the parameters are set in the parameter changing unit 8 in advance, they may be configured so that the user can input them from the quality condition input unit 12.

In the above example, the recognition result evaluation unit 6
Was evaluated using the character recognition accuracy of the characters of the entire form, but the same evaluation was performed for characters in a predetermined area on the form or in an area set from the quality condition input unit 12 or the like. May go. Also, contrary to the above eligibility condition, "the number of characters whose character recognition accuracy of the first candidate is equal to or less than a predetermined threshold among all characters of the recognized document text or characters included in a predetermined portion is equal to or smaller than a predetermined Below the reference ratio. "

Also, all the characters of the recognized document text,
Alternatively, the average value of the character recognition accuracy of the first candidate for the characters included in the predetermined partial area may be determined, and the eligibility condition may be based on the magnitude comparison between the average value and the predetermined reference value.

In the above example, the feedback loop is
The repetition was performed until the quality was qualified or the image reading parameters were restricted. In addition to these conditions, or instead of the latter among these conditions, the number of evaluations in the recognition result evaluation unit 6 is a predetermined number. It is also possible to configure so that the loop is stopped by reaching the upper limit value, and the document text data at that time is output to the recognition result output unit 14. If the original document image is poor, the quality may not improve even if the image reading parameters are adjusted. In such a case, as described above, by setting an upper limit on the number of evaluations, avoiding useless loop processing in which quality is not improved is effective from the viewpoint of reducing processing time and improving throughput. .

The parameter changing section 8 may adjust a plurality of image reading parameters. For example,
The combination of the resolution and the density threshold value may be changed appropriately and the image input control unit 10 may be instructed. At that time, the order of change between the parameters can be given a priority. For example, at a certain resolution, the quality can be evaluated by changing the density threshold in several steps within a predetermined width, and if the quality does not improve, a parameter changing method of increasing the resolution by one step can be adopted.

The image input unit 2 may have a scaling function for generating a reduced image. In this case,
The reading magnification of the scanner can be used as the image reading parameter. For example, in the first time, the document image is read with the magnification set to a small value (for example, reduced to 60%). Then, only when sufficient quality cannot be obtained at this magnification, the magnification is increased and the document image data is generated again, and the character recognition and evaluation are performed.
As described above, by increasing the magnification only when necessary, the average size of the document image data is suppressed, and the character recognition processing can be speeded up.

In the above-described example, the image reading parameters are adjusted by feedback control for each form. However, quality evaluation and parameter adjustment are performed only on the first one or a few sheets, thereby making it possible to obtain desired values. If a parameter value that achieves the quality of is found, the parameter value is fixed at that value, and in subsequent forms, quality evaluation,
A configuration in which feedback control for parameter adjustment is not performed is possible. This is because, among a plurality of forms included in a set of documents, conditions such as font size and noise level are easy to recognize when the form is a printed matter, and when handwritten characters are used. This is because the level of whether or not is almost the same in many cases. This makes it possible to quickly complete the process of reading a document composed of a large number of forms.

Although the above-described apparatus has been described by way of example having an automatic paper feed mechanism, the present invention is not premised on that. An essential feature of the present invention is that a single form is read. It is also demonstrated in ensuring quality.

[0058]

According to the document reading apparatus of the present invention, the recognition result evaluation unit evaluates the reading quality of the document text data based on the character recognition accuracy output by the character recognition unit.
If the quality does not satisfy the predetermined eligibility condition, the parameter changing unit changes the image reading parameters. That is, the read document text data is objectively and quantitatively evaluated, and the evaluation result is fed back to the image input unit, whereby the effect of improving the reading quality is obtained. Since this feedback does not involve the user, the effect of reducing the burden on the user in reading the document and improving the efficiency is obtained, and the effect of ensuring the reliability of the generated document text data is realized. You.

When the reading resolution is used as the image reading parameter, the initial value is set to a predetermined coarse resolution, and only when the desired quality cannot be obtained,
A higher resolution is set, so that the time required for reading is reduced on average, and the area is reduced when, for example, storing image files by suppressing the image file size on average. This is particularly convenient when reading a large number of documents.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a document reading apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic diagram illustrating an example of a character recognition process performed by a character recognition unit.

FIG. 3 is a processing flowchart illustrating processing in the present apparatus.

[Explanation of symbols]

2 image input unit, 4 character recognition unit, 6 recognition result evaluation unit, 8 parameter change unit, 10 image input control unit, 1
2 Quality condition input unit, 14 recognition result output unit, 20 documents.

Claims (8)

[Claims] 1. A document reading apparatus comprising: an image input unit that reads a document image and generates document image data; and a character recognition unit that performs character recognition processing on the document image data and generates document text data. A recognition result evaluation unit that evaluates the reading quality of the document text data based on the character recognition accuracy that the character recognition unit evaluates and outputs for each character, and determines whether the evaluation result satisfies a predetermined eligibility condition. And a parameter changing unit that changes an image reading parameter in the image input unit when the eligibility condition is not satisfied. 2. The method according to claim 1, wherein the image input unit generates the document image data for the same document again based on the changed image reading parameters, and outputs the generated document image data to the character recognition unit. Item 1. The document reading device according to Item 1. 3. The document reading apparatus according to claim 1, wherein the image reading parameter includes a resolution for reading the document image. 4. The document reading apparatus according to claim 1, wherein the image reading parameter includes a density threshold in reading the document image. 5. The document reading apparatus according to claim 1, wherein the image reading parameter includes a resolution for reading the document image, and the parameter changing unit increases the resolution. . 6. A parameter initializing unit that sets a predetermined initial value as the image reading parameter in the first reading of the document, wherein the parameter initializing unit sets a predetermined coarse resolution as an initial value of the resolution. 6. The document reading device according to claim 5, wherein: 7. The recognition result evaluating unit calculates a character recognition accuracy average value obtained by averaging the character recognition accuracy for each character included in an entire region or a predetermined partial region of a document, and calculates the character recognition accuracy average value and a predetermined reference value. 2. The document reading apparatus according to claim 1, wherein the eligibility of the evaluation result is determined based on a comparison with a value. 8. The recognition result evaluating unit obtains a ratio of characters having the character recognition accuracy within a predetermined range among characters included in an entire region or a predetermined partial region of a document, and compares the ratio with a predetermined reference ratio. 2. The document reading apparatus according to claim 1, wherein the eligibility of the evaluation result is determined based on the evaluation result.