JP2001084333A - Character reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character reader of a high recognition rate. SOLUTION: Character information entered or the like on a paper sheet medium 9 is optically read in an image scanner part 8 and the image data Din are stored in a data buffer 2 as two-dimensional image data Dxy. A character area detection part 3 divides them into the image data Dc of respective character units and outputs them. The data conversion parts (a1)-(an) of plural systems (n) affine-transform the image data Dc and generate rotary image data P1-Pn rotated to respectively prescribed rotation angles. The feature extraction parts b1-bn of the plural systems (n) similarly extract feature extraction data α1-αn from the rotary image data P1-Pn and a judgment part 4 obtains a parameter (estimated value) ωclosest to a character corresponding to the image data Dc by executing a signal processing by a maximum likelihood estimation method to the feature extraction data α1-αn. Then, a collation part 5 reads character code data corresponding to the estimated value ω from a character code storage part 7 and outputs them.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reader for automatically recognizing characters written or printed on a medium such as paper.

[0002]

2. Description of the Related Art In a conventional character reading apparatus, characters written or printed on a paper medium are read by optical means, and two-dimensional image data obtained thereby is subjected to signal processing to characterize the characters. Has been extracted. Furthermore, the reference feature data corresponding to each of the various types of character code data prepared in advance is compared with the feature data obtained by the feature extraction, and the character code data having the most similarity is recognized as a recognition result. It is to be output.

[0003]

With the recent development of electronic equipment to which a microcomputer or the like is applied,
It is extremely important to manage and operate character information as digital data. For this reason, if the field of application of the character reading device only corresponds to a specific limited document, the response to electronic devices becomes insufficient, and it is necessary to fill in general documents such as newspapers and magazines and paper media of any form. Alternatively, reading printed character information at a higher recognition rate has been a major issue.

[0004] Further, if signal processing is performed on two-dimensional image data read by conventional optical means to extract character features and perform precise arithmetic processing in order to increase the recognition rate, the processing takes a long time. Because of the necessity, there has been a problem that convenience cannot be provided to the user.

In particular, when character information on a paper medium is read by an optical means, if the character information is read while being tilted (rotated) with respect to the optical means, a tilt correction operation is performed on the two-dimensional image data. Although the recognition rate is increased by performing the processing, there is a problem that this arithmetic processing takes an extremely long time.

SUMMARY OF THE INVENTION The present invention has been made to overcome such problems, and an object of the present invention is to provide a character reading apparatus capable of improving the character recognition rate and shortening the processing time.

[0007]

According to the present invention, there is provided a character reading apparatus for automatically recognizing character information written or printed on a medium, wherein the character reading device detects the character information as image data. Means, data conversion means for generating rotation image data by performing rotation conversion processing on the image data detected by the detection means at a predetermined rotation angle, and characteristics of the rotation image data generated by the data conversion means A characteristic extracting means for extracting the character information based on the characteristic data generated by the characteristic extracting means by statistical signal processing. The extracting means is constituted by a plurality of processing systems composed of at least two or more combinations, and the determining means comprises an upper part of the plurality of processing systems. And identifies the character information based on the feature data generated by the plurality of feature extraction means.

According to this configuration, the detecting means detects the character information written on the medium and outputs the image information as image data. The image data having the character information is supplied in parallel to respective data rotating units in a plurality of processing systems. Each data rotator performs a conversion process for rotating the image data at a specific rotation angle, and supplies each rotation image data generated by this conversion process to a corresponding feature extractor. Each feature extracting unit extracts the feature of each of the rotated image data and outputs the respective feature data to the determining unit in parallel. The determination means specifies character information by performing statistical signal processing on the plurality of feature data supplied in parallel.

As described above, high-speed character recognition is made possible by providing a plurality of data conversion means and feature extraction means and performing parallel processing.

In addition, a plurality of feature data are generated from a plurality of rotation image data rotated by a specific rotation angle, respectively, and character information is specified by performing statistical signal processing on the feature data having different rotation angle parameters. Therefore, even when the detection means detects character information on the medium when the character information is tilted (rotated) with respect to the detection means, an error factor caused by the tilt is automatically corrected,
Improve the identification accuracy (recognition accuracy) of character information.

Further, in the character reading device, the character code data stored in the storage means is collated with the storage means for storing the character code data in advance, and the character code data stored in the storage means is collated based on the specified result of the determination means. It is characterized by comprising a collating means for outputting character code data corresponding to the information.

With this configuration, the character code data corresponding to the character information specified by the determining means is output from the comparing means. By supplying this character code data to a computer, a communication device, or the like, it is possible to support digital electronic devices.

[0013]

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 main part of the character reading device according to the present embodiment. FIG. 2 is an explanatory diagram schematically showing functions of data conversion units a1 to an.
5 is a flowchart for explaining the operation of the character reading device.

In FIG. 1, the character reading apparatus includes a system controller 1, a data buffer 2, a character area detecting section 3, a determining section 4, a collating section 5, a reference data storing section 6, a character code storing section 7, an image scanner section. 8, multiple n
(N is an arbitrary number of 2 or more) data conversion units a1 to an,
And a plurality of n feature extraction units b1 to bn. Here, the character region detection unit 3 and the determination unit 4, the collation unit 5, the data conversion units a1 to an, and the feature extraction units b1 to b
n is a digital signal processor (Digital Signal Processor: operating under the control of the system controller 1).
DSP).

The image scanner unit 8 includes a line sensor (not shown) in which a number of fine light receiving elements are arranged along the sub-scanning direction x, and a direction orthogonal to the sub-scanning direction x (hereinafter, main scanning). The sheet medium 9 supplied from the direction y) is optically read.

For example, when the user operates the image scanner unit 8
When the image scanner unit 8 and the sheet medium 9 are moved in opposite directions along the main scanning direction y, the line sensor of the image scanner unit 8 causes the line sensor of the image scanner unit 8 to move. By repeating the main scanning reading and the sub scanning reading, character information written on the paper medium 9 is optically read two-dimensionally. Although not shown, the two-dimensional optical reading can also be performed by fixing the image scanner unit 8 and automatically conveying the paper medium 9 to the image scanner unit 8 by an automatic paper feed mechanism. Will be

The system controller 1 controls a reading operation of the image scanner unit 8 by executing a preset computer program, and has a built-in microprocessor (MPU) for controlling the operation of the entire character reading apparatus. Have been.

The data buffer 2 is formed of a semiconductor memory such as a DRAM (Dynamic RAM) which can be erased and re-stored, and pixel data Din output line-sequentially from the image scanner unit 8 at the time of the above optical reading. Is input and stored as two-dimensional dot matrix data (hereinafter, referred to as image data) Dxy in frame units.

The character area detecting section 3 identifies pixel data corresponding to an individual character portion and pixel data corresponding to a background portion in the image data Dxy stored in the data buffer 2, and Region (range) is specified. That is, the gradation level of each pixel data constituting the image data Dxy is compared with a predetermined threshold value, and pixel data having a gradation level smaller than the predetermined threshold value is set as a background, and a pixel having a gradation level larger than the predetermined threshold value is set as a background. The data is identified as a character part, and a relatively large area of pixel data corresponding to the character part is determined as an individual character area (font size), and the image data Dxy is divided for each area. .
Then, the divided individual image data Dc is output as individual character information. The individual image data Dc is divided into rectangular shapes based on an orthogonal coordinate system corresponding to the sub-scanning direction x and the main scanning direction y.

The data converters a1 to an input the image data Dc supplied from the character area detector 3 in accordance with an instruction from the system controller 1, and further convert the input image data Dc into predetermined rotation angles θ1 to θn, respectively. Is performed to perform rotation, and 2 obtained by the calculation is performed.
Dimensional image data (hereinafter referred to as rotated image data) P1
~ Pn.

Referring to FIG. 2 in more detail, the data converter a1 outputs image data D as shown in FIG.
By rotating c based on the rotation angle θ1,
It generates and outputs rotated image data P1 as shown in FIG. The data conversion unit a2 generates and outputs rotated image data P2 as shown in FIG. 3C by performing a rotation process on the same image data Dc based on the rotation angle θ2.

Similarly, the data conversion units a3 to an also
By rotating the image data Dc based on the rotation angles θ3 to θn, rotated image data P3 to Pn are generated and output as shown in FIGS.

Here, these rotation angles θ1 to θn are set to different values from each other, and specific values of the rotation angles θ1 to θn are determined by a look-up table (not shown) provided in the system controller 1. ) Is stored in advance.

Further, in this embodiment, the data converter a1
Is usually set to θ1 = 0 °, and the rotation angles θ2 to θn of the other data conversion units a2 to an
Are set to a plus angle and a minus angle by substantially the same number with respect to the rotation angle θ1.

Further, in the present embodiment, an affine transformation (af
fine transformation) is used. However, the present invention is not limited to the affine transformation, and another method may be used.

The feature extraction units b1 to bn are provided by a data conversion unit a
By performing a feature extraction process on the rotated image data P1 to Pn generated by 1 to an, each rotated image data P1
ＰPn are extracted, and the feature extraction data α1 to αn obtained by the feature extraction process are output to the determination unit 3.

That is, each of the feature extraction units b1 to bn is connected to each of the data conversion units a1 to an so that n processing systems are constituted by the data conversion units a1 to an and the feature extraction units b1 to bn. Have been. The data conversion units a1 to an and the feature extraction units b1 to bn perform parallel processing in synchronization with a predetermined timing, so that the feature extraction data α1 to αn can be obtained from the input of the image data Dc in these n parallel systems. Are almost equal to each other.

In this embodiment, each of the feature extraction units b1 to b1
bn is calculated using the feature extraction data α1 to α1
n is generated.

That is, according to the pattern matching method,
The standard template data of a large number of character images stored in advance in a storage unit (not shown) provided in a part of the feature extraction units b1 to bn and the reference data storage unit 5 is compared with each of the rotated image data P1 to Pn. Then, by extracting a part where the standard template data and each rotated image data coincide with each other as a constituent part of the character, the feature extraction data α1 to α
Generate n.

According to the structural analysis method, each rotated image data P
1 to Pn, data corresponding to the line element of the character is extracted and analyzed, and the feature extraction data α1 to αn are obtained by using the end points, concave portions, convex portions, and loop portions of the character, and changes in their size and shape as characteristic parameters. Generate

The judgment unit 3 uses the maximum likelihood estimation method to extract feature extraction data α1 supplied in parallel from the feature extraction units b1 to bn.
To αn and standard feature data stored in the reference data storage unit 6 in advance, and a parameter ω that maximizes the likelihood function (simultaneous probability density) is set as a parameter ω to be recognized as a recognition target character (image data (The character of Dc).

The reference data storage section 6 is formed of a read-only memory (ROM), and stores standard characters (JIS first).
Standard feature data corresponding to the level, characters conforming to the JIS second level, etc.) are stored in advance. That is, a large number of feature data corresponding to each of a large number of standard characters is stored, and when the determination unit 3 performs the above-described determination processing, the determination unit 3 supplies these characteristic data to the determination unit 3.

The character code storage unit 7 is formed of a read-only memory (ROM), and stores character code data corresponding to the standard characters in advance.

The collation unit 5 refers to the character code storage unit 7 based on the parameter (estimated value) ω data supplied from the determination unit 3 and outputs the character code data corresponding to the data as recognition result data. I do.

Next, an example of the operation of the character reading apparatus having such a configuration will be described with reference to the flowchart of FIG.

In FIG. 3, when the user turns on the main power supply or the like, the character reading apparatus starts its operation and waits until the paper medium 9 is supplied to the image scanner section 8 (step 100). Next, when the paper medium 9 is supplied to the image scanner unit 8, the transport mechanism or the like starts operating to automatically transport (automatically feed) the paper medium 9, and accompanying the transport, the image scanner unit 8 Reads the character information written on the paper medium 9 and stores the read image data Din in the data buffer 2 (step 102). When the end of the paper medium 9 passes through the image scanner section 8, the reading is completed (step 1).
04).

Next, in step 106, the character area detecting section 3 divides the image data Dxy stored in the data buffer 2 into individual character areas, and divides the first image data Dc into data conversion sections a1. Output to ~ an.

Next, in step 108, the data converters a1 to an set different rotation angles θ1 to θn in accordance with instructions from the system controller 1, respectively.
Next, in step 110, the image data Dc is subjected to affine transformation processing at different rotation angles θ1 to θn to generate rotated image data P1 to Pn as shown in FIG.

Next, the feature extraction units b1 to bn perform feature extraction on each of the rotated image data P1 to Pn, thereby generating feature extraction data α1 to αn (step 1).
12) Next, the determination unit 4 determines the estimated value closest to the character from the feature extraction data α1 to αn by the maximum likelihood estimation method (step 114).

Next, in step 116, it is determined whether or not the character can be specified based on the estimated value. If the character cannot be specified, the rotation angles θ1 to θn are finely adjusted (step 122). , The processing from step 110 is repeated.

On the other hand, if it is determined in step 116 that the character can be specified, the process proceeds to step 118, where the collating unit 5 reads the character code data corresponding to the estimated value from the character code storage unit 7. Output.

Next, at step 124, it is determined whether or not the recognition processing for all the character image data stored in the data buffer 2 has been completed. The recognition processing ends.

On the other hand, if it is determined in step 124 that the processing has not been completed, the process proceeds to step 126, in which the character area detecting section 3 converts the image data Dc corresponding to the next character into the data conversion sections a1 to an. After the transfer, the processing from step 108 is repeated.

As described above, according to the character reader of this embodiment, the data conversion units a1 to an and the feature extraction units b1 to
Since bn is provided in a plurality of systems and parallel processing is performed, high-speed character recognition becomes possible. Further, a plurality of pieces of rotated image data P1 rotated by specific rotation angles θ1 to θn, respectively.
To generate a plurality of feature extraction data α1 to αn from
Character information is specified (recognized) by performing arithmetic processing on these feature extraction data α1 to αn having different parameters of the rotation angles θ1 to θn by the maximum likelihood estimation method. When detecting the character information on the medium 9, the character information is tilted (rotated).
Even in the case of detection, the error factor caused by the inclination can be automatically corrected, and the recognition accuracy of the character information can be improved.

In the present embodiment, the maximum likelihood estimation method is used as the statistical signal processing. However, the present invention is not limited to this, and a statistical signal processing other than the maximum likelihood estimation method is applied. May be.

[0046]

As described above, according to the present invention, a plurality of data conversion means and feature extraction means are provided for parallel processing, so that high-speed character recognition becomes possible. In addition, character information is recognized by generating a plurality of feature data from a plurality of rotation image data rotated at specific rotation angles, respectively, and performing statistical signal processing on the feature data having different rotation angle parameters. Even when the detecting means detects the character information on the medium while the character information is tilted (rotated) with respect to the detecting means, the error factor caused by the inclination is automatically corrected, and the character information recognition accuracy is improved. Can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a character reading device according to an embodiment.

FIG. 2 is an explanatory diagram schematically showing a function of a data conversion unit.

FIG. 3 is a flowchart for explaining the operation of the character reading device according to the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... System controller 2 ... Data buffer 3 ... Character area detection part 4 ... Judgment part 5 ... Collation part 6 ... Collation data storage part 7 ... Character code storage part 8 ... Image scanner part 9 ... Paper medium a1-an ... Data conversion part b1 to bn: feature extraction unit

Claims (2)

[Claims] 1. A character reading apparatus for automatically recognizing character information written or printed on a medium, comprising: a detecting unit for detecting the character information as image data; and a detecting unit for detecting the image data detected by the detecting unit. A data conversion unit that generates rotation image data by performing a rotation conversion process at a rotation angle of: a feature extraction unit that generates a feature data by extracting a feature of the rotation image data generated by the data conversion unit; Determining means for specifying the character information based on the feature data generated by the feature extracting means by statistical signal processing; wherein the data converting means and the feature extracting means comprise a plurality of processes comprising at least two or more combinations And the determination unit is configured to output the feature data generated by the plurality of feature extraction units in the plurality of processing systems. Character reader and identifies the character information based on. 2. A storage unit for storing character code data in advance, and a character corresponding to the character information by collating the character code data stored in the storage unit based on a specified result of the determination unit. 2. The character reading device according to claim 1, further comprising a matching unit that outputs code data.