JPH06124365A - Character recognizing device - Google Patents

Abstract

PURPOSE:To enable a user of the character recognizing device to easily optimize a binarization threshold value or the read density of a scanner by actually recognizing a document. CONSTITUTION:The recognition mode is switched to a special mode by a mode switching part 11, and the document is set on the scanner 1 and recognized several times while the binarization threshold value of a character image binarization part 3 or the read density of the scanner 1 is varied by a binarization threshold value setting part 4. The optimum binarization or read density is determined on the basis of the recognition results and then set. The recognition mode is switched to a normal mode thereafter and then the best recognition performance for documents of the same quality is obtained.

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 for recognizing a binary image of a multivalued character image input from a scanner or a binary character image directly input from a scanner.

[0002]

2. Description of the Related Art In general, an image processed by a character recognition device is a black-and-white binary image of a CCD sensor output of a scanner by a threshold value. This binarization is performed on the scanner side (when directly inputting a binary image from the scanner) or on the character recognition device side (when inputting a multi-valued image from the scanner). In order to obtain good recognition performance, The scanner reading density (in the former case) or the binarization threshold value (in the latter case) must be optimized according to the difference in the original density.

Regarding such optimization of the read density or the binarization threshold value, various methods as described below have been proposed.

For example, there are the modal method and the differential histogram method shown on page 67 of "Introduction to Computer Image Processing" (Hideyuki Tamura, Soken Shuppan, 1985). The modal method is to obtain a histogram of density values of a given image and use the density values of the valleys of the two peaks as a binarization threshold value. Since the differential histogram method is considered to have a sudden change in density value at the boundary between the object and the background in the image,
This is a method of determining a threshold value by using a differential value (rate of change in density) instead of directly using the density value of an image.

Further, there is a method shown in "Method of deciding threshold value from density distribution" (Otsuno Nobuyuki) on page 145 of "National Congress of Information and Information Society of 1993". this is,
This is a method of determining the optimum binarization threshold value based on integration using only the 0th and 1st moments of the concentration distribution.

Further, there is an optimum binarization method disclosed in Japanese Patent Publication No. 60-37952. This is to store a multi-valued video signal in a video buffer, binarize the video signal read from the video buffer with different slice levels by a variable slice circuit, and obtain each of the obtained binarized video signals. The line width amplification factor (number of black dots / number of surroundings) is obtained, and the slice level (2) of the slice circuit is calculated based on the obtained plurality of line width amplification factors and the reference line width amplification factor.
(Value threshold).

[0007]

However, in the mode method, it is difficult to determine the threshold value in the case of a document having a poor printing condition in which a clear valley does not occur in the histogram. The differential histogram method does not work effectively for an image in which the density value near the boundary between the object and the background changes intricately. The threshold value determination method based on the density distribution is not effective as a process for crushing or blurring a "line" as an image to be handled in pattern recognition such as character recognition. JP-A-60-37952
In the optimal binarization method of the signal, the optimal binarization threshold may not be obtained depending on the recognition algorithm.

As described above, the above-mentioned conventional technique is greatly affected by the quality of the input image and the recognition algorithm, and it is difficult to determine an appropriate binarization threshold value or reading density for various originals.

Therefore, as one idea, in the character recognition device, character recognition is actually performed by variously changing the binarization threshold value or the reading density, and the optimum binarization threshold value or the reading density is obtained from the obtained recognition result. A method of determining is conceivable.

However, in a recent general character recognition apparatus, it is not always possible to accurately judge the suitability of the binarization threshold value or the read density from the recognition result. This is because the appropriateness of the binarization threshold value or the read density, or the density state or quality of the binary character image to be recognized is not reflected as it is in the recognition result.

That is, in a recent character recognition device, it is common to perform post-processing such as language (word) processing after recognition of each character by pattern matching to obtain a final recognition result. is there. Even when the candidate rank of the correct character is lowered in the recognition result by the pattern matching because the binarization threshold value or the read density is not optimal (when it is erroneously recognized), this is corrected by the post-processing such as language processing. Then, the user cannot recognize from the final recognition result that the binarization threshold value or the read density is inappropriate.

Therefore, an object of the present invention is to enable a character recognition device to accurately determine an optimum binarization threshold value or reading density for various documents based on the result of actual character recognition by a user. It is to be.

[0013]

According to the first aspect of the present invention, character recognition for recognizing a character image obtained by binarizing a multi-valued character image input from a scanner or a binary character image directly input from a scanner. In addition to the recognition mode for normal character recognition (normal mode), the device is provided with a recognition mode (special mode) for determining the binarization threshold value of a multi-valued character image or the reading density of the scanner. A means for selecting two kinds of recognition modes and a means for variably setting the binarization threshold value of the multi-valued character image or the reading density of the scanner are provided, and the special mode includes processing that is not directly affected by the density state of the character image. No

According to the second aspect of the present invention, in the normal mode, the feature amount of the binary character image and the feature amount registered in the pattern dictionary are collated, and the recognition result obtained by the collation is subjected to language processing or the like. The post-processing is performed to obtain the final recognition result. In the special mode, the final recognition result is obtained by matching the feature amount of the binary character image with the feature amount registered in the pattern dictionary. To be done.

According to the third aspect of the present invention, in the special mode, a dedicated pattern dictionary in which the characteristic quantities of a plurality of images of the same character having different density states are registered by assigning different character codes to each other is used. .

[0016]

According to the character recognition device of the first, second or third aspect of the invention, the special mode does not include the process (the language process in the second aspect of the invention) which is not directly influenced by the density state of the character image. When the character recognition is actually performed, the suitability of the binarization threshold value or the read density is reflected in the recognition result more accurately than in the normal mode. Therefore, the user
Easily by recognizing the character image of the document you want to recognize or another document of the same print density with different binarization thresholds or reading densities, and judging whether the binarization threshold or reading density is appropriate from the recognition result. The optimum binarization threshold value or read density can be determined.

According to the character recognition device of the invention of claim 3,
In the special mode, when an image of a character registered in the pattern dictionary dedicated to this mode is input and recognized, a binarization threshold or a character code that differs depending on the read density is obtained as a recognition result. It becomes extremely easy to determine whether the read density is appropriate and to determine the optimum binarization threshold value or read density.

[0018]

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

Embodiment 1 FIG. 1 shows a schematic structure of a character recognition device. In the character recognition apparatus according to the present embodiment, the scanner 1 scans a document, and the character image input unit 2 captures multivalued image data of the document from the scanner 1. This multi-valued image data is converted into 2 by the character image binarization unit 3.
The value is converted and input to the character cutout unit 5. The binarization threshold of the character image binarization unit 3 is variably set by the binarization threshold setting unit 4. The user can change the binarization threshold through the binarization threshold setting unit 4 by operating a panel (not shown).

Binary image data is directly input from the scanner 1, and the binary image data is directly input to the character cutout unit 5 without passing through the character image binarization unit 3 to obtain a binarization threshold value. The setting unit 4 can also variably set the reading density of the scanner 1 (see broken lines A and B).

The character cutout unit 5 cuts out individual character image data from the input binary image data and inputs it to the character recognition unit 6. The character recognition unit 6 normalizes the input character image data and then extracts the feature amount, and collates the feature amount with the feature amount registered in the pattern dictionary 7,
Obtain the recognition result (distance between candidate character and dictionary, etc.). The language processing unit 8 performs the language processing (word matching, etc.) using the language knowledge (for example, words) in the language dictionary 9 as post-processing on the recognition result input from the character recognition unit 6 to obtain the final recognition result. Is sent to the recognition result output unit 10.

The flow of such recognition processing is in the case of a recognition mode (normal mode) for normal character recognition.
Apart from this, there is a recognition mode (special mode) for determining the binarization threshold of the character image binarization unit 3 (or the read density of the scanner 1). In the special mode, the character recognition unit 6
The recognition result obtained by is output to the recognition result output unit 10 as a final recognition result. That is, the recognition processing in the special mode does not include the language processing that is not directly influenced by the density state of the recognition target character / character image.

Reference numeral 11 is a mode switching unit for switching the recognition mode to the normal mode or the special mode. The user
The recognition mode can be switched through the mode switching unit 11 by operating a panel unit (not shown).

When the user wants to optimize the binarization threshold value or the read density, the user switches the recognition mode to the special mode. Then, a document of the quality (paper quality, background color, character color, etc.) assumed by the user is set on the scanner 1, and the binarization threshold setting unit 4 recognizes while changing the binarization threshold or the reading density. The process is repeatedly executed, and the binarization threshold value or the read density at which the recognition rate of the recognition result output from the recognition result output unit 10 is the highest is obtained as the optimum value.

If the optimum binarization threshold value or read density thus determined is set through the binarization threshold value setting unit 4 and the recognition mode is switched to the normal mode to perform character recognition, the highest quality will be obtained for a document of equivalent quality. The recognition performance of can be obtained.

In the special mode, the influence of the density state (quality) of the character image on the recognition rate, that is, the recognition rate of the character recognition section 6 varies considerably depending on the character type. An example thereof is shown in FIG.

As can be seen from FIG. 2, the recognition rate of "Kai" changes greatly depending on the difference in the binarization threshold value or the read density (difference in the density state of the character image), and the recognition rate becomes the highest. The binarization threshold or read density range is small. On the other hand, the recognition rate peaks for "call" and "yui" spread, and the recognition rate peaks for "day" and "present" are not fixed. It is impossible to determine the optimum binarization threshold value or read density.

Therefore, test originals of various qualities on which character types such as "Kira" which directly reflects the density state of the character image are directly reflected in the recognition result are prepared, and the binarization threshold value or the read density is determined. In this case, it is preferable to select and use a test document that is close to the quality of the document that the user wants to recognize.

Embodiment 2 FIG. 3 shows a schematic structure of a character recognition device. As is clear from a comparison between FIG. 1 and FIG. 3, in the present embodiment, the pattern dictionary 12 dedicated to the special mode is added. In the special mode, the pattern dictionary 12 is replaced by the pattern dictionary 7 and the character recognition unit. Used in 6.

FIG. 4 shows a pattern dictionary 12 dedicated to the special mode.
It is a conceptual diagram which shows the content of. In this example, the characteristic amount 13 of a character image having a good character “amount”, the characteristic amount 14 of a crushed (darkly read) character image, and the characteristic amount 15 of a faint (lightly read) character image are registered. Further, the character codes of “amount”, “dark”, and “light” are assigned to the respective feature amounts.

For the purpose of correcting the variation in the quality of the scanner 1 at the time of introducing the character recognition device and subsequently correcting the change in the read density due to the deterioration of the scanner 1, for example, as shown in FIG. Prepare a standard document on which "amount" is printed at standard density. Then, the mode is switched to the special mode, and the standard document is read by the scanner 1 for recognition.

For example, when the recognition result as shown in FIG. 6A is obtained, it may be judged that the current binarization threshold value or the read density is optimum. However, if the binarization threshold value is too low (the reading density is too dark), the feature amount 14 can be matched by the comparison with the special mode dedicated pattern dictionary 12, and the character code of “dark” in FIG. Since such a recognition result is obtained, the binarization threshold value is raised (or the reading density is lowered) and recognition is performed again to confirm the recognition result. On the other hand, if the binarization threshold is too high (the reading density is too light), the recognition result as shown in FIG. 6C including the “light” character code can be obtained. Therefore, lower the binarization threshold (reading). Re-recognize by increasing the density) and check the recognition result.

When the user wants to set the optimum binarization threshold value or reading density for the original document to be recognized, an original document having a quality similar to that of the original document shown in FIG. 5 is prepared and used. It suffices to recognize the special mode and make adjustments.

In the case of this embodiment, different character codes are output depending on whether the binarization threshold value or the read density is appropriate. Therefore, by automatically controlling the binarization threshold value setting unit 4 based on the recognition result. It is also easy to automate the adjustment of the binarization threshold value or the read density.

[0035]

As can be understood from the above description, according to the inventions of claims 1 to 3, the user of the character recognition apparatus can easily set the binarization threshold value or the read density according to the document to be recognized. Since the optimum recognition performance can be achieved by optimizing, and in particular, according to the invention of claim 3, whether the binarization threshold value or the read density is appropriate, too large, or too small appears in the recognition result as a difference in character code. This makes it extremely easy to determine whether the binarization threshold value or the read density is appropriate and optimize it.

[Brief description of drawings]

FIG. 1 shows a device configuration of a first embodiment.

FIG. 2 shows a relationship between a binarization threshold value (reading density) and a recognition rate.

FIG. 3 shows a device configuration of a second embodiment.

FIG. 4 shows the contents of a special mode dedicated pattern dictionary.

FIG. 5 shows an example of a document used in a special mode.

FIG. 6A shows a recognition result when the binarization threshold value is appropriate. (B) An example of the recognition result when the binarization threshold is low is shown. (D) An example of the recognition result when the binarization threshold is high is shown.

[Explanation of symbols]

 1 Scanner 2 Character Image Input Section 3 Character Image Binarization Section 4 Binarization Threshold Setting Section 5 Character Extraction Section 6 Character Recognition Section 7 Pattern Dictionary 8 Language Processing Section 9 Language Dictionary 10 Recognition Result Output Section 11 Mode Switching Section 12 Special Mode-specific pattern dictionary

[Procedure amendment]

[Submission date] October 19, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Figure 6

[Correction method] Change

[Correction content]

FIG. 6A shows a recognition result when the binarization threshold value is appropriate. (B) An example of the recognition result when the binarization threshold is low is shown. (C) An example of the recognition result when the binarization threshold is high is shown.

Claims (3)

[Claims] 1. A character recognition device for recognizing a character image obtained by binarizing a multivalued character image input from a scanner or a binary character image directly input from a scanner, and a recognition mode for normal character recognition. , A binarization threshold value of a multi-valued character image or a recognition mode for determining the reading density of the scanner, a means for selecting these two types of recognition modes, and a binarization threshold value of the multi-valued character image or a scanner And a means for variably setting the reading density of the character recognition device, and the latter recognition mode does not include processing that is not directly affected by the density state of the character image. 2. The character recognition device according to claim 1, wherein
In the recognition mode for normal character recognition, the feature amount of the binary character image is compared with the feature amount registered in the pattern dictionary,
Post-processing such as language processing is performed on the recognition result obtained by the collation to obtain the final recognition result. In the recognition mode for determining the binarization threshold value or the read density, the feature amount of the binary character image is obtained. A character recognition device characterized in that a final recognition result is obtained by collating with a feature amount registered in a pattern dictionary. 3. The character recognition device according to claim 2,
In the recognition mode for determining the binarization threshold value or the read density, use a dedicated pattern dictionary in which the characteristic quantities of a plurality of images of the same character with different density states are assigned different character codes and registered. Character recognition device.