JP2000155806A - Character recognition method and its device and dictionary preparation method and its device and character quality judgment method and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quickly and highly precisely attain character recognition from a high quality character to a low quality character even in a multifont environment. SOLUTION: A character quality judgment parameter detecting part 103 detects a character quality judgment parameter including the 'variance of the maximum value of a distance conversion image' of an input character image, and a condition judging part 104 judges a condition related with the parameter so that whether or not the input character image is 'high quality' or 'low quality' can be judged. A dictionary selecting part 108 selects a dictionary 112 for a high quality character or a dictionary 114 for a low quality character according to the judged character quality, and a matching part 107 performs the character matching of the selected dictionary with the input character image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a character recognition technique.

[0002]

2. Description of the Related Art Character recognition uses a characteristic amount of a character image, but the characteristic amount changes considerably depending on the quality of the character image. An example of such a character recognition technique in consideration of the quality of a character image is described in Japanese Patent Application Laid-Open No. 7-57057. In this conventional technique, a plurality of masks, which are features of the same character pattern, are stored in a dictionary, and the input image and the dictionary are matched to determine the Nth recognition candidate category, the distance value, and the dictionary mask number. Then, the quality of the character image is determined based on these, and candidate categories are selected according to the character quality.

As a general method for determining the quality of a character image, a method using a character line width is known.

[0004]

SUMMARY OF THE INVENTION The above-mentioned Japanese Patent Application Laid-Open No. 7-570.
In the technology of No. 57, the quality of a character is determined after matching a feature amount extracted from a character image with a dictionary. However, since a plurality of masks of different categories are stored in the dictionary, character quality determination is performed due to the influence between categories. There is a concern that the accuracy may decrease. Further, since a plurality of masks for the same character pattern are stored in the dictionary, the dictionary becomes large. However, matching with a large dictionary takes time, which may cause a reduction in the overall processing speed.

A character quality judgment method using character line width can be expected to have a certain degree of accuracy in an environment where only one type of font is used. High precision judgment cannot be expected in the environment.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a character recognition method and apparatus capable of accurately and quickly recognizing high-quality character images to low-quality character images even in a multi-font environment. . Another object of the present invention is to provide a method and apparatus for creating a dictionary used in such a character recognition device. Another object of the present invention is to provide a character quality determination method suitable for a multi-font character recognition device and the like.

[0007]

According to the character recognition method of the present invention, a character image input step of inputting a character image to be recognized, and a character for judging the quality of the character image from the input character image in the character image input step. A parameter detection step of detecting a quality determination parameter; a condition determination step of determining the quality of the input character image by performing a condition determination on the character quality determination parameter detected by the parameter detection step; A dictionary selecting step of selecting a high-quality character dictionary prepared in advance when determined, and selecting a low-quality character dictionary prepared in advance when determined to be low quality in the condition determining step; A character recognition step of performing character recognition of the input character image using the dictionary selected in the step; Outputting a result recognized by the character recognition step, wherein the character quality determination parameter detected in the parameter detection step includes a variance of a maximum value of a distance conversion image of the input character image. And According to a preferred aspect of the present invention, the ratio of the number of black pixels of the input character image to all pixels and the number of black pixels of the input character image are divided by the contour length to the character quality determination parameter detected in the parameter detection step. And a value obtained by dividing the contour length of the input character image by the number of black pixels having the maximum value of the distance-converted image of the input character image.

The character recognition device of the present invention detects a character image input unit for inputting a character image to be recognized, and a character quality judgment parameter for judging the quality from the character image input by the character image input unit. A character quality determination unit that determines the quality of the input character image by performing a condition determination regarding the detected character quality parameter, a low-quality character dictionary, a high-quality character dictionary, and is determined to be high quality by the character quality determination unit. A character recognition unit that performs character recognition of the power character image using the low-quality character dictionary when the quality is determined to be low, and the character recognition unit. And a character quality determination parameter detected by the character quality determination unit includes a variance of a maximum value of a distance-converted image of the input character image. And butterflies.

According to the dictionary creation method of the present invention, a character image input step of inputting a character image for creating a dictionary, and a character quality judgment parameter for judging the quality is detected from the input character image in the character image input step. A parameter detecting step of performing a condition determination on the quality of the input character image by performing a condition determination on the character quality determination parameter detected by the parameter detecting step; and the input determined to be high quality by the condition determining step. Using the image as a learning image for a high-quality character dictionary, creating the high-quality character dictionary; and converting the input image determined as low-quality by the condition determining step to a low-quality character dictionary learning image. A dictionary creating step of creating the dictionary for low-quality characters by utilizing The character quality determining parameters detected by flops, characterized in that includes the variance of the maximum values of the distance transform image of the input character image.

A dictionary creation apparatus according to the present invention detects a character image input unit for inputting a character image for creating a dictionary, and a character quality judgment parameter for judging the quality from a character image input by the character image input unit. A character quality determination unit that determines the quality of the input character image by performing a condition determination on the detected character quality parameter; and converts the input image determined to be high quality by the character quality determination unit into a high-quality character. The high-quality character dictionary is created using the learning image of the dictionary for the high-quality character, and the input image determined to be low-quality by the character quality determination unit is used as the learning image of the low-quality character dictionary to generate the low-quality character dictionary. The character quality determination parameter detected by the character quality determination unit includes a variance of the maximum value of the distance conversion image of the input character image. The features.

The character quality determination method according to the present invention is characterized in that the variance of the maximum value of the distance-converted image of the character image is used as a parameter for determining the quality of the character image.

[0012]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a character recognition / dictionary creating apparatus according to an embodiment of the present invention. This character recognition / dictionary creation device has both a function as a character recognition device and a function to create a dictionary for character recognition, and is configured to also serve as a common functional part.

In FIG. 1, a character image input unit 100 includes:
For example, a character image to be recognized or a learning character image for creating a dictionary is input from an image input device such as a scanner, a storage device, or a communication line. The character recognition determination unit 102 is a unit that determines the quality of the character image input from the character image input unit 100. The character image determination unit 102 detects a character quality determination parameter of the input character image.
And a condition determining unit 104 that performs a condition determination of the detected character quality determination parameter to determine whether the input character image is “high quality” or “low quality”. This determination result is sent to the character recognition unit 106 and the dictionary creation unit 110.

The character recognition unit 106 is a high quality character dictionary 1
This is a part for performing character recognition of an input character image using the 12 or the low-quality character dictionary 114. This character recognition unit 106
Is a high-quality character dictionary 112 or a low-quality character dictionary 1 according to the character quality determined by the character quality determination unit 102.
And a matching unit 10 for performing feature matching between the selected dictionary and the input character image.
And 7. The output unit 116 includes the matching unit 1
07 is a unit for outputting the recognition result to an output device such as a printer or a display, a medium such as a storage device or a floppy disk, or a communication line.

The dictionary creation unit 110 includes the character image input unit 10
This is a part for creating a high-quality character dictionary 112 and a low-quality character dictionary 114 by using a character image input by 0 as a learning image for creating a dictionary. Character quality judgment unit 10
2 is used as a learning image for the high-quality character dictionary 112, and the input character image determined to be low-quality is used as a learning image for the low-quality character dictionary 114. Is done.

The processing will be described more specifically. First,
The processing content of the character recognition device will be described with reference to the flowchart shown in FIG.

In FIG. 2, a character image to be recognized is input from the character image input unit 100 (step 200). Character quality determination parameter detection unit 1 of character quality determination unit 102
03, one or more types of character quality determination parameters are detected from the input character image (steps 202 to 20).
7).

First, detection of an essential character quality determination parameter will be described. A distance conversion image is created from the input character image (step 202). This distance-converted image is an image in which each black pixel in the character image is labeled with the distance to the closest white pixel. FIG. 3A shows an example of a distance conversion image of a high-quality character image of the character “I”. Next, the maximum value of the distance conversion image is calculated (step 203). The maximum value is the maximum value of the label attached to the black pixel of the distance conversion image. FIG. 3B shows an image representing the maximum value of the distance conversion image shown in FIG. Such a maximal value image is also called a skeleton in the technical field related to the present invention. Next, the variance of the maximum value of the distance conversion image is calculated (step 20).
4). The dispersion of the maximum value is an essential character quality determination parameter.

The quality of a character image can be determined with higher accuracy in various types of fonts, as compared with the method using the character line width, using only the essential character quality determination parameters. That is, multi-fonts can be handled. This will be described with reference to FIG. FIG. 8 shows character images of various types of “I” for five fonts.
The results of sorting the distance-converted images in descending order of the variance of the maximum value are shown. From FIG. 8, it can be confirmed that the character images are sorted in order from the crushed character image to the blurred character image according to the quality for any font type. This means that the variance of the maximum value of the distance conversion image is extremely excellent as a character quality determination parameter even in a multi-font environment.

As described above, it is possible to determine the character quality with high accuracy by using only the variance of the maximum value of the distance conversion image.
In order to further improve the determination accuracy, the following three types of parameters are also detected. However, whether or not to use these three types of parameters and which one or more parameters to use can be specified by the user as necessary.

As one of the parameters, the ratio of the black pixels of the input character image to all the pixels is calculated (step 205). As another parameter, a value obtained by dividing the number of black pixels of the input character image by the contour length is also calculated (step 2).
06). Here, the contour length (more precisely, the black pixel contour length)
Is the number of black pixels adjacent to white pixels. As another parameter, a value obtained by dividing the contour length by the number of black pixels having the maximum value of the distance-converted image is calculated (step 207).

Condition determining section 104 of character quality determining section 102
In the above, a condition determination is performed on the character quality determination parameters detected as described above, and it is determined whether the input character image is “high quality” or “low quality” (step 210). Table 1 below shows an example of this determination condition.

[0023]

[Table 1] That is, the condition “variance of the maximum value of the distance conversion image” is larger than the threshold value 1. Condition "Ratio of black pixels to all pixels" is threshold value 0.4
Greater than. Condition “The value obtained by dividing the number of black pixels by the contour length” is larger than threshold value 2. The condition “value obtained by dividing the contour length by the number of black pixels having the maximum value of the distance-converted image” is larger than the threshold value 1.5. Are determined as "low quality" when all the conditions are satisfied, and as "high quality" when at least one condition is not satisfied. However, only the conditions relating to the parameters to be used are to be determined. Therefore, for example, when only the essential “variance of the maximum value of the distance-converted image” is used as the character quality determination parameter, the character quality is determined only based on whether or not the condition is satisfied. In addition, when the “ratio of black pixels to all pixels” is used, “low quality” is determined only when the conditions are satisfied at the same time, and “high quality” is determined otherwise. .

FIG. 4A shows a distance-converted image of a crushed (low-quality) character image of the character "I", and FIG. 4B shows an image of the maximum value. Table 2 below shows examples of character quality judgment parameters relating to this character image and the high-quality character image of the character “I” corresponding to the distance conversion image shown in FIG.

[0025]

[Table 2] As is apparent from a comparison between Table 2 and Table 1, in the character image corresponding to FIG. 3A, each parameter is smaller than the threshold value. Therefore, all of the above conditions are not satisfied, and the result is determined to be “high quality”. In the character image corresponding to FIG. 4A, each parameter is larger than the threshold value, and therefore, any of the above conditions is satisfied, and it is determined that the quality is "low quality".

Next, the dictionary selection unit 10 of the character recognition unit 106
In step 8, if the input character image is determined to be “low quality”, the low quality character dictionary 114 is selected (step 21).
1) If it is determined to be “high quality”, the high quality character dictionary 112 is selected (step 213). The matching unit 107 determines the recognition result by performing feature matching between the dictionary (114 or 112) selected by the dictionary selection unit 108 and the input character image (step 21).
3). It should be noted that after a plurality of candidates are obtained by feature matching, post-processing such as knowledge processing may be further performed to determine the final recognition result. The recognition result is output by the output unit 116 (Step 214).

As described above, according to the present invention, even in a multi-font environment, the quality of a character image is determined with high accuracy, and two types of dictionaries (112 or 114) are selectively used depending on the determined character quality. Since each dictionary is prepared exclusively for recognition of high-quality characters or low-quality characters, both high-quality character images and low-quality character images can be recognized with high accuracy. The size of each dictionary is
It can generally be smaller than the size of a dictionary built to accommodate all character qualities. Therefore, when character recognition is performed by selecting a dedicated dictionary corresponding to the character quality as in the present invention, the processing speed can be increased as compared with a configuration using a common large-size dictionary regardless of the character quality.

The processing contents of the character recognition / dictionary creating apparatus of the present invention as a dictionary creating apparatus will be described with reference to the flowchart shown in FIG.

In FIG. 5, first, a character image input unit 100
Input a character image as a learning image for creating a dictionary (step 300). The character quality judgment parameter is detected from the input character image by the character quality judgment parameter detection unit 103 of the character quality judgment unit 102 (step 301). The contents of this detection processing and the contents of the character quality judgment parameters to be detected are shown in steps 202 to 20 in FIG.
7 is as described above. The condition determination for the detected character quality determination parameter is performed by the character quality determination unit 10.
The second condition determination unit 104 determines whether the input character image is "high quality" or "low quality" (step 30).
2). The details of this condition determination are as described in connection with step 210 in FIG. If the input character image is determined to be a low-quality character, the dictionary creation unit 110 creates a low-quality character dictionary 114 by using the input character image as a learning image for the low-quality character (step 303). If it is determined to be high quality, the high-quality character dictionary 112 is created using the input character image as a learning image for high-quality characters (step 304).

As described above, the character quality of a character image can be determined with high accuracy. Then, a dictionary is created by selectively using the input character image as a learning image for high-quality characters or a learning image for low-quality characters according to the determination result of character quality. Therefore, a high-quality character dictionary 112 that enables high-precision recognition of high-quality characters and a low-quality character dictionary 114 that enables high-precision recognition of low-quality characters can be created.

The above-described method or apparatus of the present invention comprises:
For example, as shown in FIG.
Input interface 402, output interface 4
03, auxiliary storage device 404, floppy disk or CD
A computer-readable recording medium 4 such as a ROM
It can also be realized by software on a general computer having a configuration in which a drive 405 for 06 and the like are connected by a bus 407. In this case, a program for causing a computer to execute each processing step as described with reference to FIG. 2 or FIG. 5 is, for example, read from a recording medium 406 on which the processing is recorded into the memory 401 via the drive 405. Or the auxiliary storage device 40
4 and read into the memory 401 at the time of execution. The high-quality character dictionary and the low-quality character dictionary are stored in, for example, the auxiliary storage device 404, and are stored in the memory 400 when the process is performed.
Is read in. A character image to be recognized or a character image as a learning image for creating a dictionary is read into the memory 402 by a scanner connected via the input interface 402 or read into the memory 402 from the auxiliary storage device 404, for example. Or the recording medium 40
6 is read into the memory 402. The recognition result is output to, for example, an output device connected via the output interface 403, to the auxiliary storage device 404, or to the recording medium 406.

Further, the method or apparatus of the present invention can be realized as a client / server system as schematically shown in FIG. For example, on the server computer 500 side, a program for causing the character quality determination unit 102, the character recognition unit 106, and the dictionary creation unit 110 of FIG. , 502, a program for executing the processing corresponding to the character image input unit 100 and / or the output unit 116 is provided, the character images are input from the client computers 501 and 502 to the server computer 500, and the character recognition result is obtained. From the server computer 500 to the client computer 501,
A configuration in which the data is output to 502 is possible.

[0033]

As described above, according to the character recognition method and apparatus of the present invention, even in a multi-font environment, both high-quality character images and low-quality character images can be processed at high speed and high quality. Accurate character recognition becomes possible. According to the dictionary creation method and apparatus of the present invention, a high-quality character dictionary and a low-quality character dictionary for enabling such high-speed and high-precision character recognition can be created. Further, according to the character quality determination method of the present invention, it is possible to obtain the effects of enabling the quality of a character image to be determined with high accuracy even in a multi-font environment.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a character recognition / dictionary creating apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing flow as a character recognition device.

FIG. 3 is a diagram illustrating an example of a distance-converted image of a high-quality character image of the character “I” and its maximum value image.

FIG. 4 is a diagram illustrating an example of a distance-converted image of a crushed character image of the character “I” and its maximum value image.

FIG. 5 is a flowchart showing a processing flow as a dictionary creation device.

FIG. 6 is a block diagram illustrating a configuration example of a computer for realizing the present invention by software.

FIG. 7 is a schematic diagram of a client-server system for realizing the present invention by software.

FIG. 8 is a diagram showing a result of sorting character images of various quality “I” with respect to five types of fonts in descending order of the variance of the maximum value of the distance conversion image.

[Explanation of symbols]

 Reference Signs List 100 Character image input unit 102 Character quality judgment unit 103 Character quality judgment parameter detection unit 104 Condition judgment unit 106 Character recognition unit 107 Matching unit 108 Dictionary selection unit 110 Dictionary creation unit 112 High quality character dictionary 114 Low quality character dictionary 116 Output Department

Claims (9)

[Claims] 1. A character image input step of inputting a character image to be recognized, a parameter detection step of detecting a character quality determination parameter for determining the quality of the character image input in the character image input step, A condition determination step of determining the quality of the input character image by performing a condition determination on the character quality determination parameter detected by the detection step; a high-quality character prepared in advance when the condition determination step determines that the quality is high; A dictionary for selecting a low-quality character dictionary prepared in advance when a low-quality dictionary is determined by the condition determining step; and inputting the input characters using the dictionary selected in the dictionary selecting step. A character recognition step of performing character recognition of an image, and a character recognition step performed by the character recognition step. And a character quality determination parameter detected in the parameter detecting step includes a variance of a maximum value of a distance-converted image of the input character image. 2. The character recognition method according to claim 1, wherein the character quality determination parameter detected in the parameter detecting step further includes a ratio of black pixels of the input character image to all pixels. 3. The character recognition according to claim 1, wherein the character quality judgment parameter detected in the parameter detection step further includes a value obtained by dividing the number of black pixels of the input character image by a contour length. Method. 4. A value obtained by dividing the contour length of the input character image by the number of black pixels having the maximum value of the distance-converted image of the input character image to the character quality determination parameter detected in the parameter detecting step. The character recognition method according to claim 1, wherein the character recognition method is included. 5. A character image input unit for inputting a character image to be recognized, a character quality judgment parameter for judging the quality of the character image input from the character image input unit, and the detected character quality parameter A character quality determination unit that determines the quality of the input character image by performing a condition determination on the dictionary, a low-quality character dictionary, a high-quality character dictionary,
When the character quality determination unit determines that the quality is high, the high-quality character dictionary is used. When the quality is low, the low-quality character dictionary is used. A character recognizing unit that performs the recognition, and an output unit that outputs a recognition result by the character recognizing unit. The character recognition device characterized by including. 6. A character image input step of inputting a character image for creating a dictionary, a parameter detecting step of detecting a character quality determination parameter for determining the quality of the character image input in the character image input step, A condition determination step of determining the quality of the input character image by performing a condition determination on the character quality determination parameter detected by the parameter detection step; and converting the input image determined to be high quality by the condition determination step into a high quality character image. Creating the high-quality character dictionary by using it as a learning image of the dictionary; and using the input image determined to be low-quality by the condition determining step as a learning image of the low-quality character dictionary. A dictionary creation step for creating a quality character dictionary, wherein the characters detected by the parameter detection step are A method for creating a dictionary, wherein the quality determination parameter includes a variance of a maximum value of a distance-converted image of the input character image. 7. A character image input unit for inputting a character image for creating a dictionary, a character quality determination parameter for determining the quality of the character image input from the character image input unit by the character image input unit, and the detected character A character quality determining unit that determines the quality of the input character image by performing a condition determination regarding the quality parameter; and the input image determined to be high quality by the character quality determining unit as a learning image for a high quality character dictionary. The high quality character dictionary is created using the input image determined as low quality by the character quality determination unit as a learning image of the low quality character dictionary to create the low quality character dictionary. A dictionary creation unit, wherein the character quality determination parameter detected by the character quality determination unit includes a variance of a maximum value of a distance-converted image of the input character image. Place. 8. A character quality judging method, wherein a variance of a local maximum value of a distance-converted image of a character image is used as a parameter for judging the quality of the character image. 9. A program for causing a computer to execute a character image input step, a parameter detection step, a condition determination step, a dictionary selection step, a character recognition step, and a recognition result output step according to claim 1, 2, 3, or 4. A computer-readable recording medium characterized by being recorded.