JP3226355B2 - Recognition result evaluation method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device,
In particular, the present invention relates to a method for evaluating the certainty of a recognition result of a character recognition device that performs character recognition of a Japanese document including English characters and symbols.

[0002]

2. Description of the Related Art In a character recognition apparatus, it is practically impossible to achieve a recognition rate of 100% when a Japanese document is targeted. Therefore, it is necessary for a person to check and correct an error in the recognition result. There is. In order to make such correction work more efficient, the likelihood of the recognition result is evaluated, and the information is displayed on the display together with the character of the recognition result (for example, the display color of the character of the recognition result is changed according to the certainty). Switching) is known.

As a premise of performing such output control of the recognition result, it is necessary to evaluate the certainty of the recognition result. As the method, there is a method of evaluating the likelihood based on the similarity or the difference (distance) of the first candidate character obtained by dictionary matching at the time of character recognition, and the difference between the first candidate character and the second candidate character. A method of evaluating the likelihood by using the method (Japanese Patent Laid-Open No. 62-280983) is known.

[0004]

However, the similarity or difference obtained by dictionary matching greatly varies with the same character due to differences in the character recognition method (algorithm), the quality of the original image, the font, and the like. However, conventional methods for evaluating certainty lack versatility and stability. For this reason, conventionally, it has been troublesome to adjust a threshold value for evaluating the similarity or the difference degree according to the recognition method of the character recognition device or the target document.

An object of the present invention is to provide a new recognition result evaluation method which solves such a problem.

[0006]

SUMMARY OF THE INVENTION Recognition result evaluation according to the present invention
Value method, in the character recognition apparatus, when a plurality of candidate characters as a recognition result is obtained, with the number of strokes or configuration line primes the plurality of candidate characters, evaluate the likelihood of the recognition result
The main point is to do. More specifically,
The difference between the maximum stroke count and the minimum stroke count of multiple candidate characters
Therefore, the difference value is used for determining the certainty of the recognition result. Ah
Alternatively, the maximum number of constituent line elements of multiple candidate characters
Find the difference value from the minimum number of constituent line elements, and use the difference value to confirm the recognition result.
Used for determination of mustache.

[0007]

The constituent line prime number is the total number of pseudo straight lines when a character is represented as a set of pseudo straight lines, and generally has a larger value than the number of strokes (the number of strokes of the kanji "mouth" is three). Is 4), which better represents the structural features of the character.

[0008] The number of strokes and the number of constituent line elements are numerical values that are uniquely determined for each character, and are not affected by differences in the character recognition method, original image quality, font, and the like. Therefore, according to the method of the present invention, even if the character recognition method or the like is different, no special adjustment is required, and the reliability of the recognition result can be stabilized.
High-precision evaluation becomes possible.

[0009]

Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1 FIG. 1 is a block diagram of a character recognition device according to this embodiment. In FIG. 1, an image input unit 100 reads an image of a document using an optical scanner or the like, and stores the image data in an image memory 102. Line / character extraction unit 104
Cuts out a character line and a character image to be recognized from the image memory 102 and stores the data of the cut out character image in the character image memory 106. The character recognizing unit 108 reads the data of the character image from the character image memory 106, performs pre-processing such as noise removal and normalization, and then extracts a characteristic amount. The characteristic amount is stored in the character dictionary memory 110. A plurality of candidate characters having a large degree of similarity (a small degree of difference) are determined by comparing with the feature amount, and if necessary, the candidate characters are truncated, and the candidate characters are stored in the recognition result memory 112. The truncation process is performed by an appropriate method according to the recognition algorithm, such as a method of cutting the absolute value of the similarity as a threshold, a method of cutting the distance difference between candidates having adjacent ranks as a threshold, a method of outputting all possible candidates, and the like. I just need.

In this embodiment, the stroke number information of the characters within the recognition target range is stored in the stroke number information memory 114 in order to use the stroke number of the candidate character in evaluating the likelihood of the recognition result. The stroke number information may be stored in the character dictionary memory 110. The likelihood determining unit 116 recognizes the recognition result memory 1
The stroke number information is acquired by referring to the stroke number information memory 114 based on the candidate character code stored in 12, the likelihood of each recognition result is determined based on this, and the result is stored in the recognition result memory 112. The details of the certainty determination processing will be described later.

The result output unit 118 stores the recognition result memory 11
The content (No. 2 candidate character code, certainty, etc.) is output to an output device such as a display or a printer. For example, the first candidate character of the recognition result is added with a symbol, modified such as shading, or changed in display color or luminance according to the likelihood, and output. Such output control allows the user to easily search for a character that requires correction.

Next, the processing contents of the likelihood determining section 116 will be described. FIG. 2 is a flowchart of the process.

First, it is checked whether the number of candidates for one noted recognition result in the recognition result memory 112 is two or more characters (step 200). If the candidate is one character, a predetermined fixed value is output as a certainty value of the recognition result (step 215).

If the candidate has two or more characters, the first
The candidate characters are read (step 210), the number of strokes is obtained by referring to the stroke number information memory 114 based on the character code (step 220), and if there are remaining candidates (step 225), the next candidate character is read (step 220). Step 2
30), the number of strokes is obtained from the stroke count information memory 114 (step 220). When the number of strokes has been acquired up to the last candidate, the maximum and minimum strokes are found among the acquired strokes,
The difference value is obtained (step 235). Here, all candidates are targeted, but the number of targeted candidates may be limited, for example, up to the top three candidates.

Next, the likelihood is calculated from the difference value of the number of strokes according to the following equation: likelihood = 100− (number of stroke difference × fixed value) (1) (step 240).

That is, among a plurality of candidates for the recognition result,
When there are characters with greatly different numbers of strokes, the likelihood of the recognition result is evaluated as "low", and conversely, when the difference in the number of strokes of the characters listed as candidates is small, the likelihood is evaluated as "high" . It is assumed that, when a character “quantity” is recognized using a characteristic amount such as a direction of a contour of a character image, a recognition result as shown in Table 1 is obtained. Such a recognition result tends to be obtained when the character quality of the “quantity” is poor and tends to be crushed. The character recognition method may be any method that can obtain a plurality of candidates for the input image, such as a structural analysis method or a neural network.

[0018]

[Table 1]

For each candidate character of this recognition result, Table 2
Are obtained.

[0020]

[Table 2]

Here, when targeting up to the third place candidate,
Since the maximum number of strokes is 12 and the minimum number of strokes is 2, if the “fixed value” in equation (1) is set to 5, the certainty is 50 (the larger the value, the higher the certainty).

The formula for calculating the certainty is not limited to the formula (1). Although the certainty is calculated here based on the difference between the maximum number of images and the minimum number of images, the present invention is not limited to this. For example, a method of calculating certainty using a set of a difference value and an average value between the maximum number of strokes and the minimum number of strokes,
It is also possible to adopt a method of calculating certainty using the set of the difference value between the maximum and minimum strokes and the maximum number of strokes, or the method of calculating certainty using the set of the difference value between the maximum and minimum strokes and the minimum number of strokes. It is.

Embodiment 2 FIG. 3 is a block diagram of a character recognition apparatus according to the present embodiment. The same reference numerals as in FIG. 1 denote the same parts.

In this embodiment, the calculation of the certainty determination in the first embodiment is changed to a table table. Therefore, the certainty value corresponding to the difference between the maximum number of strokes and the minimum number of strokes (formula (1)) The likelihood table memory 120 which stores the value calculated by the above) is added. Table 3 shows an example of the contents of the probability table memory 120.

[0025]

[Table 3]

FIG. 4 is a flowchart of the processing of the likelihood determining section 116A. The difference from the likelihood determination processing of the first embodiment (FIG. 2) is that instead of calculating the likelihood using the difference value between the maximum number of pictures and the minimum number of pictures obtained in step 235, the likelihood is determined by the difference value in step 240A. By referring to the likelihood table memory 120, a value of likelihood is obtained. Other than this, it is the same as the first embodiment.

For example, in the case of a recognition result as shown in Table 1, the difference between the maximum number of images and the minimum number of images is 10 from Table 2, and the value of certainty is 50 from Table 3.

As described in the first embodiment, the likelihood determining section 116A sets the difference value between the maximum number of strokes and the minimum number of strokes and the average value, the difference value between the maximum number of strokes and the minimum number of strokes, and the maximum number of strokes. Alternatively, a set of a difference value between the maximum number of images and the minimum number of images and a minimum number of images may be obtained, and a certainty value corresponding thereto may be obtained from the certainty table memory 120.

Embodiment 3 FIG. 5 is a block diagram of a character recognition apparatus according to this embodiment, and the same reference numerals as those in FIG. 1 indicate the same parts. In the present embodiment, the likelihood determining unit 116B uses the “elementary line number” instead of the “number of strokes” for the calculation of the likelihood. It is stored in the memory 124. However, the constituent line prime number information may be stored in the character dictionary memory 110.

Since the character recognition and the output of the recognition result are the same as those in the above-described embodiments, the description will be omitted, and the process of determining the certainty will be described. FIG. 6 is a flowchart of the process.

First, it is checked whether or not the number of candidates for one noted recognition result in the recognition result memory 112 is two or more characters (step 300). If the candidate is a single character, a predetermined fixed value is output as a certainty value of the recognition result (step 315).

If the candidate has two or more characters, the first
The candidate character is read (step 310), the constituent line prime number is acquired by referring to the constituent line prime information memory 122 based on the character code (step 320), and if the candidate remains (step 325), the next candidate The character is read (step 330), and the number of strokes is stored in the constituent line element information memory 1.
22 (step 320). When the constituent element numbers are obtained up to the last candidate, the maximum constituent element number and the minimum constituent element number are found among the obtained constituent element numbers, and their difference values are obtained (step 335). Here, all candidates are targeted, but the number of targeted candidates may be limited, for example, up to the top three candidates.

Next, the likelihood is calculated from the difference value of the constituent line prime numbers according to the following equation: likelihood = 100− (component line number difference value × fixed value) (2) (step 340).

That is, among a plurality of candidates of the recognition result,
When there are characters with greatly different constituent line prime numbers, the likelihood of the recognition result is evaluated as "low", and conversely, when the difference in the constituent line prime numbers of the candidate characters is small, the likelihood is evaluated as "high". Will be.

It is assumed that the constituent line numbers as shown in Table 4 are obtained for each candidate of the recognition result shown in Table 1 above.

[0036]

[Table 4]

Here, when targeting the third candidate,
Since the maximum number of constituent line elements is 14 and the minimum number of constituent line elements is 2, if the “fixed value” in the equation (2) is 5, the likelihood is 60 (the larger the value, the higher the likelihood).

The formula for calculating the likelihood is not limited to the formula (2). In addition, although the certainty is calculated here based on the difference between the maximum number of constituent line elements and the minimum number of constituent line elements, the present invention is not limited to this. For example, a method of calculating certainty using a set of a difference value and an average value between the maximum number of constituent element numbers and the minimum constituent element number, using a set of a difference value between the maximum constituent element number and the minimum constituent element number and a maximum constituent element number It is also possible to employ a method of calculating certainty, a method of calculating certainty using a set of a difference value between the maximum number of constituent line elements and the minimum number of constituent element numbers, and a set of minimum constituent element numbers.

Embodiment 4 FIG. 7 is a block diagram of a character recognition apparatus according to this embodiment, and the same reference numerals as in FIG. 5 indicate the same parts.

In the present embodiment, the calculation of the certainty determination in the third embodiment is changed to a table table, and for that purpose, the certainty value (corresponding to the difference value between the maximum number of constituent line elements and the minimum number of constituent line elements) is calculated. The likelihood table memory 124 storing the value calculated by the above equation (2) is added. Table 5 shows an example of the contents of the certainty table memory 124.

[0041]

[Table 5]

FIG. 8 is a flowchart of the process of the likelihood determining section 116C. The difference from the likelihood determination processing of the third embodiment (FIG. 6) is that instead of calculating the likelihood using the difference value between the maximum number of constituent line elements and the minimum number of constituent line elements determined in step 335, the likelihood is calculated in step 340A. By referring to the certainty table memory 124 based on the difference value, the value of the certainty is obtained. Except for this, it is the same as the third embodiment.

For example, in the case of a recognition result as shown in Table 1, the difference between the maximum number of constituent elements and the minimum number of constituent elements is 12 according to Table 4, and the certainty value is 54 according to Table 5.

As described in the third embodiment, the likelihood determining unit 116C sets a set of a difference value and an average value between the maximum number of constituent element numbers and the minimum constituent element number, and sets the maximum constituent element number and the minimum constituent element number. A set of a difference value and a maximum constituent line element number or a difference value between a maximum constituent line element number and a minimum constituent line element number and a minimum constituent line element number may be obtained, and a certainty value corresponding thereto may be obtained from the certainty table memory 124. Good.

As described above, the main feature of the present invention is to use the number of strokes or the number of constituent lines for evaluating the likelihood of the recognition result. The evaluation value is combined with the table value by another method. It is also possible to determine the final certainty.

[0046]

As described above in detail, the method of the present invention uses a stroke number or a constituent line prime number, which is a numerical value uniquely determined for each character, to determine the likelihood of the recognition result. It is possible to stably and highly accurately evaluate the certainty of the recognition result without requiring any special adjustment even if the document image quality and the document image quality and fonts are different.

[Brief description of the drawings]

FIG. 1 is a block diagram of a character recognition device according to a first embodiment.

FIG. 2 is a flowchart of a recognition result likelihood determining process according to the first embodiment.

FIG. 3 is a block diagram of a character recognition device according to a second embodiment.

FIG. 4 is a flowchart of a recognition result likelihood determination process according to the second embodiment.

FIG. 5 is a block diagram of a character recognition device according to a third embodiment.

FIG. 6 is a flowchart of a recognition result likelihood determination process according to the third embodiment.

FIG. 7 is a block diagram of a character recognition device according to a fourth embodiment.

FIG. 8 is a flowchart of a recognition result likelihood determination process according to the fourth embodiment.

[Explanation of symbols]

REFERENCE SIGNS LIST 100 Image input unit 102 Image memory 104 Line / character cutout unit 106 Character image memory 108 Character recognition unit 110 Character dictionary 112 Recognition result memory 114 Stroke number information memory 116, 116A, 116B, 116C Probability determining unit 118 Result output unit 120, 124 Probability table memory 122 Constituent line prime number information memory

Claims (4)

(57) [Claims] 1. A character recognition apparatus, when a plurality of candidate characters as a recognition result is obtained, of the plurality of candidate ideographic
A recognition result evaluation method characterized by determining the likelihood of the recognition result based on the number of large strokes and the minimum number of strokes . 2. The recognition result according to claim 1 , wherein a difference value between a maximum stroke number and a minimum stroke number of the plurality of candidate characters of the recognition result is obtained, and the difference value is used to determine the certainty of the recognition result. Evaluation method. 3. A character recognition apparatus, when a plurality of candidate characters as a recognition result is obtained, of the plurality of candidate ideographic
A recognition result evaluation method, comprising determining the likelihood of the recognition result based on the number of large constituent line elements and the minimum number of constituent line elements. 4. The method according to claim 3 , wherein a difference value between a maximum number of constituent line elements and a minimum number of constituent line elements of a plurality of candidate characters of the recognition result is obtained, and the difference value is used to determine the certainty of the recognition result. Recognition result evaluation method described.