KR100957508B1 - System and method for recognizing optical characters - Google Patents

Abstract

Disclosed are an optical character recognition system and method. The optical character recognition system includes a document format setting unit, an image registration unit, a character recognition unit, and a data storage unit. The document format setting unit sets a format of a document to be recognized and a recognition area to perform character recognition in the document, and the image register unit inputs a document image to perform character recognition. In addition, when the document image corresponds to the document format set in the document format setting unit, the character recognition unit recognizes a character in the recognition area by comparing with a reference value generated by character learning, and the data storage unit recognizes the document format setting unit and the image. Stores data generated by the register and character recognition unit. When any document inputted in such a configuration corresponds to a document format in which a recognition area is set, it is possible to quickly recognize only a necessary portion to increase the efficiency of character recognition and enable character recognition in large quantities.

OCR, optical character recognition, recognition area

Description

System and method for recognizing optical characters

The present invention relates to an optical character recognition system and method. More particularly, the present invention relates to an optical character recognition system and method for recognizing only an area set in an input document by presetting an area for character recognition.

Recently, the importance of archiving documents and documents is increasing. Therefore, the documents and documents that were managed in the form of paper are being digitalized for convenient management. In order to electronically convert the documents in paper, people have to see the documents and manually enter them manually.

Today, to reduce this hassle, digitalization is achieved by scanning documents or documents and extracting characters from the scanned images using optical character recognition technology (OCR).

Optical Character Recognition (OCR) refers to a technology for recognizing the content of a portion of a document image input through a scanner.

However, there is a problem in character recognition using optical character recognition, because in general, optical character recognition is mainly implemented in the form of reading the entire document to recognize the entire document. That is, after character recognition through optical character recognition, there is a problem in that it is necessary to separately classify necessary parts and unnecessary parts of the entire recognized document.

The present invention has been made to solve such a conventional problem, and when any input document corresponds to a document format in which a recognition area is set, only necessary portions can be quickly recognized, thereby improving the efficiency of character recognition and in large quantities. It is an object of the present invention to provide an optical character recognition system and method for enabling character recognition.

In order to achieve the above object, the optical character recognition system according to the present invention includes a document format setting unit, an image registration unit, a character recognition unit, and a data storage unit.

The document format setting unit sets a format of a document to be recognized and a recognition area to perform character recognition in the document, and the image register unit inputs a document image to perform character recognition.

In addition, when the document image corresponds to the document format set in the document format setting unit, the character recognition unit recognizes a character in the recognition area by comparing with a reference value generated by character learning, and the data storage unit recognizes the document format setting unit and the image. Stores data generated by the register and character recognition unit.

When any document inputted in such a configuration corresponds to a document format in which a recognition area is set, it is possible to quickly recognize only a necessary portion to increase the efficiency of character recognition and enable character recognition in large quantities.

The character recognition unit adjusts the overall size of the document image and evolves the pixel values of the image to recognize the characters, thereby clearly distinguishing the characters from the background.

In addition, the character recognition unit may recognize a character by identifying the line by connecting the lowest pixels among pixels having a pixel value indicating a character in the binarized document image.

Also, the character recognizer may recognize a character by grouping pixels adjacent to each other among pixels having a pixel value indicating a character in the binary document image.

The optical character recognition system according to the present invention may further include a character learner that learns a plurality of character images and stores them in a data storage to generate a reference value of the character.

The character learning unit can learn characters by adjusting the overall size of the image to be learned and binarizing pixel values of the image, and grouping pixels adjacent to each other among pixels having pixel values referring to characters in the binarized learning image. Can learn.

In addition, the character learning unit may normalize the grouped pixels to a predetermined size, and learn the character corresponding to the predetermined learning data in the normalized group.

The present invention also provides a document format setting step of setting and storing a format of a document for character recognition and a recognition area for character recognition in the document, an image registration step of inputting a document image for character recognition; If the document image corresponds to the format of the document set in the document format setting step, optical character recognition comprising a character recognition step of recognizing and storing the characters in the recognition area compared with the reference value generated by the character learning Provide a method.

In this case, in the optical character recognition method according to the present invention, before the character in the recognition area is compared with the reference value, the entire size of the document image is reduced to meet a preset standard, and the pixel value of the reduced image is reduced. It is preferable to include the document image conversion step of binarizing.

Further, in the optical character recognition method according to the present invention, the character recognition step has a pixel value indicating a character in the binarized image when the binarized document image corresponds to the set document format after the document image conversion step. Among the pixels, it is preferable to further include a document image line identification step of connecting the lowest pixels to identify the line.

Also, in the optical character recognition method according to the present invention, the character recognition step further includes a document character group generation step of grouping pixels adjacent to each other among pixels having a pixel value indicating a character in the binarized document image. desirable.

The optical character recognition method according to the present invention may further include a character learning step of learning and storing a plurality of character images to generate a reference value of the character before the document format setting step.

At this time, in the optical character recognition method according to the present invention, before learning a plurality of character images, the character learning step is to reduce the overall size of the image to be learned according to a preset standard, and to binarize the pixel value of the reduced image It is preferable to include a learning image conversion step.

Also, in the optical character recognition method according to the present invention, the character learning step further includes a step of generating a learning character group for grouping adjacent pixels among pixels having pixel values indicating a character in the binarized learning image. desirable.

Also, in the optical character recognition method according to the present invention, the character learning step further includes a character designation step corresponding to a predetermined learning data by normalizing the grouped pixels to a preset size after the learning character group generation step. It is desirable to.

Further, in the optical character recognition method according to the present invention, when the separation distance of two or more grouped pixels is located within a preset separation distance range, the group may be grouped into one group.

In the present invention, in the recognition of the characters written by the user in the standardized document form, the area of the standardized form is set in advance as a character recognition target area, and when the character recognition is performed, By performing character recognition only on the character recognition target area instead of performing document recognition, the character recognition speed can be improved and more accurate character recognition can be achieved, thereby improving the efficiency of the character recognition as a whole.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings. In order to more clearly understand the present invention, like reference numerals designate like elements in different drawings.

1 is a schematic block diagram of an embodiment of an optical character recognition system according to the present invention.

An optical character recognition system for recognizing only a region set in an input document image by using a document format in which an area to be recognized by text is preset is a character learning unit 100, a document format setting unit 200, and an image register unit ( 300, a character recognition unit 400, and a data storage unit 500.

The character learner 100 learns a plurality of character images and stores them in the data storage unit 500 to generate a reference value of the character. FIG. 2 is a diagram illustrating an embodiment of a text image serving as a reference value. When learning a text image, the text image of various fonts as illustrated in FIG. 2 may be input in a predefined order to perform learning. .

That is, the text image shown in FIG. 2 is input, information about pixel values of the corresponding image is extracted, the reference value is generated by the text learner 100, and stored in the data storage 500.

In this case, it is preferable to generate a reference value by extracting pixel value information of a plurality of character images having different fonts (eg, Arial, Gothic, Ming and Korean fonts, etc.), thereby increasing accuracy in character recognition. .

The document format setting unit 200 sets a format of a document to be recognized and a recognition area to perform character recognition in the document. In other words, documents with fixed forms (eg, insurance contracts, financial product contracts, etc.) have a limited portion that can be directly written by the customer. (The parts that are written directly by the customer) are set in advance.

The format and recognition area of the document may be set as a vector, and the format of the vector may be expressed as' [document name, data order, lower left coordinate, lower right coordinate, upper left coordinate, upper right coordinate]. Here the data order can start at zero.

For example, the coordinates of the location rectangle of the area to be recognized in the document 'Insurance Certificate 1' are represented as' (100, 150), (200, 150), (100, 180), (200, 180) 'and' (300, 150), (400, 150), (300, 180), (400, 180) ', the format of the vector is' [insurance certificate 1, 0, 100,150, 200,150, 100,180, 200,180]' It can be represented as '[insurance certificate 1, 1, 300, 150, 400, 150, 300, 180, 400, 180]'.

The image registration unit 300 inputs a document image to perform character recognition. The image register 300 may correspond to a device such as a scanner capable of imaging a document.

In addition, when the document image corresponds to the format of the document set by the document format setting unit 200, the character recognition unit 400 recognizes the character in the set recognition area by comparing it with a reference value generated by character learning. That is, when an arbitrary document image is input through the image register unit 300, when the document image is automatically recognized by the document format setting unit 200, the document format is automatically recognized. In other words, it recognizes the characters in the set area of the corresponding format.

Here, in order to automatically recognize the format of the document image, when setting the recognition area in the above-described document format setting unit 200, it may be preferable to include the area where the name of the document is located.

The data storage unit 500 stores data generated by the document format setting unit 200, the image registration unit 300, and the character recognition unit 400.

That is, the document format data in which the recognition area generated by the document format setting unit 200 is set, any document image data to be recognized by the image register unit 300, and the character recognition generated by the character recognition unit 400 are completed. Data can be saved.

In this case, the format of the document in which the recognition area is set may be stored in a vector format of '[insurance certificate 1, 0, 100, 150, 200, 150, 100, 180, 200, 180]' as described above.

With such a configuration, when any input document corresponds to a document format in which a recognition area is set, only necessary portions can be quickly recognized, thereby increasing the efficiency of character recognition and enabling character recognition in large quantities.

3 is a schematic flowchart of an embodiment showing a character learning step S100 and a document format setting step S200 which are preprocessing steps of a character recognition step in the optical character recognition method according to the present invention.

Character learning step (S100) first inputs a character image to learn in order to generate a reference value of the character used in character recognition (S110).

Next, the entire size of the input image is adjusted and the entire pixels of the image are binarized (S120). The overall size of the image can be adjusted to a preset standard. To reduce the size of the data, the image size is reduced by reducing only the large sized image without adjusting the size of the smaller sized image. Can be adjusted.

These resized images are binarized by dividing the entire pixel into 0 (white) or 1 (black) binaries. 0 refers to the background, 1 refers to the character, which makes it possible to clearly distinguish between the character and the background.

Next, line identification is performed for efficient character recognition in the binarized image (S130). Referring to FIG. 4, an example of general line identification is illustrated. The pixel located at the lowest of the displayed character pixels is connected and recognized as a line 1a. At this time, the upper left (2), the middle upper (3), the upper right (4), the left (5), the right (6), the lower left (7), the lower middle (8), the lower right (9) around the pixel located at the lowest By measuring the similarity of pixels located in adjacent eight directions, we can find the last point where any line ends.

The line identification step S130 may be omitted in the character learning step S100.

If a line is identified, pixels that refer to characters having pixel values of 1 adjacent to each other are grouped together (S140). In this case as well, the similarity of the pixels is measured as in the line identification step S130, and the grouping method through the similarity measurement will be described in detail later with reference to FIG. 6.

Next, the grouped training image pixels are normalized to fit the preset size, and the predetermined training data is assigned to the normalized character image (S150). For example, referring to FIG. 5, when the size of the grouped training image pixels is 10 × 20 or 30 × 40, all of the grouped training image pixels may be normalized to 20 × 30, which is a preset size. In the present invention, it is normalized to a size of 20 × 30, but may be normalized to other sizes.

As an example of designating the predetermined learning data to the normalized text, when the text image shown in Fig. 2 is learned, the designated text is' A, B, C, D,... , Z, a, b, c,... , z, 0, 1, 2,... Will be 9 '. Since the training data is fixed, the first grouping is 'A', the second grouping is 'B', and the last grouping is '9'.

In this way, a plurality of learning images specified in the learning data are integrated to generate a reference value of the character (S160).

In the present invention, any character learning method currently used may be used instead of the above-described method.

Next, the document format setting step (S200) is continued, and stores the format of the document to be recognized (S210), and then stores an area to perform character recognition in the format of the document (S220).

For example, in the case of a 'financial product agreement', the form of the document to be recognized may include information on the subscriber's personal information, the financial product name to be subscribed to, the product expiration date, the interest rate, the transfer date, and the transfer account number. Formed areas may be included.

In addition, the area in which the character recognition is to be performed may be a blank that the contractor must write in a plurality of areas constituting the above-described financial instrument contract, and stores the coordinates of the blanks.

FIG. 6 is a flowchart illustrating an embodiment of training image pixel grouping step S140 by measuring similarity described above with reference to FIG. 3.

In the training image line identification step (S130) of FIG. 3, first, all pixels of the binarized training image are set as non-visit points (S141). Then, to retrieve the pixel value of the non-visit point (S142), since there is no visited pixel at first, the first pixel of the entire image may be started.

Next, it is checked whether each pixel has a pixel value of 1, which refers to a character (S143), and if the pixel has a value of 0, which does not refer to a character, the pixel value of the non-visit point is searched again. S may retrieve the value of the next unvisited pixel, except for the previously visited pixel. If the retrieved pixel has a pixel value of 1 indicating a character, grouping starts with a new number (S144).

At this time, as shown in FIG. 4, it is checked whether the non-visited pixel has the pixel value 1 among the pixels in the eight directions (2 to 9) adjacent to the pixel having the searched 1 value (S145). If there are pixels with, the same number is grouped (S146). Then, the process is repeated from step S145 again with respect to the most recently grouped pixels.

In this case, when there are no unvisited pixels having the pixel value 1 among the adjacent 8-direction pixels, it is checked whether all pixels of the binarized learning image have been visited (S147). If there exists, the process returns to step S142 and starts again from the pixel value search of the non-visit point.

7 is a schematic flowchart of an embodiment showing a character recognition step in the optical character recognition method according to the present invention.

When the character learning step S100 and the document format setting step S200 of FIG. 3 are completed, the character recognition step is executed.

If any document image to be recognized by the character through the image registration unit 300 is input (S300), the entire size of the input document image is the same as the learning image reduction and binarization step (S120) mentioned in the above description of FIG. Reduce and binarize in a manner (S400).

When the binarized document image corresponds to the document format set in the document format setting step S200, the document image may include the learning image line identification step S130 to the learning image pixel grouping step S140 mentioned in the above description of FIG. 3. The document image lines are identified in the same manner as in (S500), and the document image pixels are grouped (S600).

Whether the document image corresponds to the set document format can be known using the format of the vector mentioned in the above description of FIG. 2. That is, if the document format is set in the vector format of [Document Name, Data Order, Lower Left Coordinate, Right Lower Coordinate, Upper Left Coordinate, Upper Right Coordinate], the coordinates of the area where 'Document Name' is set in the input document image By applying, it can be known by recognizing the characters in the area.

That is, in the present invention, after the character is first recognized through the remaining character recognition step to be described later, character recognition in the remaining setting region may be performed only on the document image corresponding to the set document format.

Next, the size of the grouped document image pixels is also normalized in the same manner as the method of normalizing the size of the grouped document image pixels in the character specifying step S150 mentioned in the above description of FIG. Let (S700).

Thereafter, among the character recognition areas set in the stored document format, text recognition is performed on the grouped document image pixels in the coordinate area corresponding to the remaining areas except for the 'document name' (S800). In this case, character recognition may be performed by comparing the reference value of the character generated in the character learning step S100 with the mean square error (MSE) value of the document image pixel to be recognized according to Equation 1 below.

The equation for obtaining MSE value is as above, T is target, the reference value of the learned character normalized to 20 × 30, I is input, and the input document image normalized to 20 × 30. Means grouped pixels.

Referring to Equation 1 in detail, the MSE value is the total number of pixels constituting one image normalized to 20 × 30 after adding up the absolute values of the target minus the pixel value of the input for each pixel. The mean value when divided.

In other words, the larger the MSE value, the larger the pixel error between the target and the input, and the MSE value is expressed as a percentage. If the value exceeds the predefined error range (for example, 25%), it corresponds to the reference value. Can be considered non-characters.

For example, referring to FIG. 8, the MSE value can be seen that since M × N = 20 × 30 = 600, the 20 × 30 image is composed of 600 pixels.

For each pixel, calculate the absolute value of the target minus the pixel value of the input: | T (x6y11) -I (x6y11) | = 1, | T (x7y12) -I (x7y12) | = 1, | T ( x8y8) -I (x8y8) | = 1, | T (x8y9) -I (x8y9) | = 1, | T (x11y8) -I (x11y8) | = 1, | T (x11y9) -I (x11y9) | = 1, | T (x14y17) -I (x14y17) | = 1, | T (x15y16) -I (x15y16) | = 1, | T (x15y20) -I (x15y20) | = 1, | T (x16y19) -I (x16y19) | = 1, | T (x16y23) -I (x16y23) | = 1, | T (x17y22) -I (x17y22) | = 1 (meaning the calculated value of the bold pixels in the image) Since the remaining pixel count is 0, the sum of absolute values is 12.

That is, if 12/600 is modified as a percentage, 2% does not exceed 25%, and thus the document image shown in FIG. 8 may be regarded as a character corresponding to the target image and may be recognized.

The input document in which character recognition is completed through the above-described steps is stored in the data storage unit 500 (S900).

The invention can also be embodied as computer readable code on a computer readable recording medium. Computer-readable recording media include all kinds of recording devices that store data that can be read by a computer system. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, which are also implemented in the form of a carrier wave (for example, transmission over the Internet). It also includes. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

1 is a schematic block diagram of an embodiment of an optical character recognition system in accordance with the present invention.

2 is a diagram illustrating an embodiment of a text image serving as a reference value.

3 is a schematic flowchart of an embodiment showing a character learning step and a document format setting step which are preprocessing steps of a character recognition step in the optical character recognition method according to the present invention;

4 illustrates an embodiment of general line identification.

5 illustrates an embodiment of normalizing the size of grouped image pixels.

6 is a flowchart illustrating an embodiment of a training image pixel grouping step using the similarity measure of FIG. 3.

7 is a schematic flowchart of an embodiment showing a character recognition step in the optical character recognition method according to the present invention.

8 is a diagram illustrating an embodiment of a target image and an input image for character recognition by comparing MSE values.

Claims (18)

A document format setting unit for setting a format of a document to be recognized for text and a recognition area for performing character recognition in the document; An image register for inputting a document image to perform character recognition; A text recognition unit for recognizing a character in the recognition area by comparing with a reference value generated by character learning, when the document image corresponds to a format of a document set by the document format setting unit; And A data storage unit for storing data generated by the document format setting unit, the image registration unit, and the character recognition unit; Including, The character recognition unit, Adjusting the overall size of the document image, binarize the pixel value of the image, and the optical recognition of the character by grouping pixels adjacent to each other among the pixels having a pixel value indicating a character in the binary document image Character recognition system. delete The method of claim 1, The character recognition unit, And recognizing a character by concatenating the lowest pixels among the pixels having a pixel value indicating a character in the binarized document image to identify a line. delete The method of claim 1, A text learning unit learning and storing a plurality of text images to generate a reference value of the text; Optical character recognition system further comprises. The method of claim 5, The character learning unit, And characterizing the image by adjusting the overall size of the image to be learned and binarizing pixel values of the image. The method of claim 6, The character learning unit, And a group of adjacent pixels among pixels having a pixel value indicating a character in the binarized learning image to learn a character. The method of claim 7, wherein The character learning unit, And normalizing the grouped pixels to a preset size, and learning the character corresponding to the predetermined training data. A document format setting step of setting and storing a format of a document to be subjected to text recognition and a recognition area to perform text recognition in the document; An image registration step of inputting a document image to be performed for character recognition; And A character recognition step of recognizing and storing a character in a recognition area in comparison with a reference value generated by character learning, when the document image corresponds to a format of a document set in the document format setting step; Including, The character recognition step, A document image conversion step of reducing the overall size of the document image to a preset standard and binarizing pixel values of the reduced image before comparing the characters in the recognition area with the reference value; And Generating a document character group grouping pixels adjacent to each other among pixels having a pixel value indicating a character in the binarized document image; Optical character recognition method comprising a. delete The method of claim 9, The character recognition step, After the document image converting step, when the binarized document image corresponds to the set document format, a document image which connects the lowest pixels among the pixels having pixel values referring to characters in the binarized image to identify a line. Line identification step; Optical character recognition method further comprises. delete The method of claim 9, A text learning step of learning and storing a plurality of text images to generate a reference value of the text before the document format setting step; Optical character recognition method further comprises. The method of claim 13, The character learning step, Before learning a plurality of text images, the learning image conversion step of reducing the overall size of the image to be learned in accordance with a predetermined standard, and binarizing the pixel value of the reduced image; Optical character recognition method comprising a. 15. The method of claim 14, The character learning step, Generating a learning character group grouping pixels adjacent to each other among pixels having a pixel value indicating a character in the binarized learning image; Optical character recognition method further comprises. The method of claim 15, The character learning step, A character designating step corresponding to a predetermined learning data by normalizing the grouped pixels to a preset size after the learning character group generating step; Optical character recognition method further comprises. The method according to claim 9 or 15, And when the separation distance of two or more grouped pixels is located within a preset separation distance range, grouping into one group. 17. A recording medium in which the optical character recognition method of any one of claims 9, 11, and 13-16 is readable by a computer, and recorded with executable program code.

Images