JPS60225983A - Reader of character - Google Patents

Abstract

PURPOSE:To attain the recognition processing of different types of characters in an optical character reader by recognizing and processing a sensor output as multiple information at reading a printed KANJI (Chinese character) and in case a hand-written KANJI, as binary information. CONSTITUTION:A photoelectric transfer analog signal obtained from a scanning system is stored as a multiple pattern in an image buffer 32 by an A/D convertor 31. Characters in the image buffer 32 are detected by a control signal 37 and character frame information of every character is transferred to a normarization part 36, which controls addresses in the buffer 32 so that characters can be normarized into the prescribed size in accordance with either the hand-written character or the printed one. In case of the printed character, a normarized character pattern 34 consisting of multiple data can be obtained. In case of the hand-written character, as for a character image, a binary pattern 35 in the prescribed normarized size is obtained by processing, in the similar manner, the multiple data of the buffer 32 and data output nearest to an extraction point. The pattern 35 is sampled and recognized.

Description

[Detailed description of the invention] [Technical field of invention] The present invention relates to an optical character reading device.

[Technical background of the invention and its problems]

Optical character reading devices (hereinafter abbreviated as OCR) can read not only printed characters but also handwritten numbers, alphabets, katakana, and even thousands of printed or handwritten kanji, and are actively commercializing them. It is progressing.

While the ability to read thousands of kanji characters is expected to lead to new developments in Japanese input, there is also a strong need for lower prices. Until now, as a Kanji recognition OCR, there are machines that only read handwritten Kanji and devices that read printed Kanji and handwritten Kanji using the same device, but there are devices that read the two types of Kanji.

Since there is great demand for both printing and handwriting, this device is developing into a particularly excellent device to meet diverse needs for kanji input.

The first technical problem in reading such printed kanji and handwritten kanji is the resolution of the optical scanning system, that is, how precisely can it be observed?

FIG. 11 explains an example of conventional OCR, in which printed kanji 10 come in various sizes and line widths.
The highest number (20 lines/mm for printing of the number used)
A certain degree of irrigation is required, and if this irrigation is made coarser,
In particular, horizontal lines may disappear. - In the case of male handwritten kanji input, there are many needs for data entry, and the address, gender base, etc. should be entered in an 8 to 9 mm square writing frame using a 0.5 mm or larger mechanical pencil, ballpoint pen, etc. I'm having it done.

Therefore, the solution If for reading handwritten kanji 11 is even if the characters are blurred, etc., it is sufficient to read 10 books/m m, which is even rougher than printed kanji, and is sufficient for reading handwritten kanji 11. It has a proven resolution for alphanumeric characters, kana characters, etc.

For these reasons, in response to a request to read printed kanji lO and handwritten kanji 11 with the same device 12, the scanning resolution is set to the high-definition resolution level of printed kanji, and when reading handwritten kanji, the scanning resolution is set to the high resolution level of 1 bit. By thinning out characters at intervals to obtain character image signals, or by adding a driving condition to OCR that print and handwritten kanji are not read at the same time.
A method is used in which the resolution of the optical system is mechanically and optically switched and the same sensor is shared.

These conventional OCRs have the disadvantage that in the former case, only half of the sensor functions when reading handwritten kanji, and in the latter case, handwritten and printed kanji cannot be read at the same time, both of which complicate the circuit structure and make the device expensive. I have to.

[Purpose of the invention]

This invention improves the drawbacks of the conventional device mentioned above.
It is an object of the present invention to provide an optical character reading device that can obtain high-quality recognition images using a sensor with a low degree of decomposition and that can be easily realized.

[Summary of the invention]

The optical character reading device uses the same sensor function to read printed kanji and handwritten kanji, and when reading printed kanji, the sensor output is recognized as multivalued information, and in the case of handwritten kanji, it is processed as binary information as before. This is an optical character reading device that performs recognition processing to switch recognition processing at high speed (electrically) even for character types with different print quality (line width) in the same device.

〔Effect of the invention〕

The resolution of the sensor unit can be increased, the number of bits of the sensor can be reduced, which helps lower costs and shortens the adjustment process.

The need for sensor resolution switching is rx < b. Electrical signal processing can be easily realized, and the structure of the optical and mechanical parts is simplified, resulting in a low-cost device. At the same time, the switching processing speed is increased, and the same form can be processed. To enable high-speed form processing even when there are no printed kanji and no handwritten kanji.

[Embodiments of the invention]

FIG. 2 explains the process of processing in an embodiment of the present invention. When inputting handwritten kanji, the scanning system resolution for handwritten kanji ll is set to 8 lines/mm, and the resulting photoelectric conversion signal is quantized as a binary signal.
5. Further, in the case of printed kanji 10, when obtaining a scanning signal with higher definition, the photoelectric conversion signal is stored as a multivalued signal when quantized, and an image thereof is shown in 26.

The preprocessing circuit switches the processing method to 211 using separately prepared control information, and converts handwritten Chinese characters and printed Chinese characters into the same number of job-specific spatial dimensions. The input pattern obtained in this way is input to the identification processing circuit 22 as a sampling pattern. Pre-prepared kanji standard pattern 2
3 is also input to the identification processing circuit 24, which performs a matching operation on the input pattern and outputs the answer.

A diagram illustrating the above-mentioned process in more detail is shown in Figure @3.
Here, 30 is a photoelectric conversion analog signal obtained from the scanning system, and 31 is stored as a multi-value pattern in an image buffer 32 using an A/D converter. Handwritten from the control unit,
Print switching control signal and inspection cut 33. Characters in the image buffer 32 are detected by the operation start signal 37 of the normalization section 34, and character frame information (character area address on the image buffer 32) for each character is transferred to the normalization section 36. The normalization unit 36 controls the address of the image buffer 32 using a control signal 37 so as to normalize it to a predetermined size depending on whether it is a handwritten character or a printed character, and in the case of a printed kanji character, the multi-valued data shown in FIG. Obtain a normalized character pattern consisting of . At this time, since printed kanji characters come in various sizes, it is not enough just to transfer the area from the image buffer 32, but needless to say, they are normalized by interpolation processing to normalize the size. Similarly, in the case of handwritten kanji, the character image read out by the address control of the image buffer 32 is preliminarily converted into a 1-bit binary pattern by neighborhood processing such as outputting the multivalued data of the image buffer 32 and the nearest neighbor data of the extraction point. The image is normalized to a predetermined normalized size and stored in 35 pattern buffers.

The character pattern normalized in this way is further
4! The sampling process shown in Figure 4 is performed. 41 in FIG. 14 is the output from the handwritten kanji normalization pattern buffer (35 in FIG. 3) and passes through a 2×2 addition mask. 42 is the output from the printing kanji regular f Hibatan buffer (FIG. 3, 34), and the selection circuit 43 selects either tl! by the handwriting/printing switching control signal 37. A window circuit 44 having a 3×3 weight addition mask consisting of a shift register is configured, and weight addition is performed on the window using 45 load tables to obtain a sampling pattern that matches the number of dimensions of the identification circuit or standard pattern. get.

In the case of printed kanji, the input to the selection circuit of 43 is the multivalued data (in this example) as is, but in the case of handwritten kanji, the values are added by a 2 x 2 mask from the shift register, and 2 The data is input to the selection circuit 43 as compressed data, and at this point the information density of the handwritten kanji and the printed kanji are the same.

FIG. 5 explains the peripheral circuit of the above-mentioned sampling circuit, in which the sampling calculation circuit 52 operates with the normalized pattern as input, stores it in the sampling buffer 53, and in response to a command from the controller 56, the identification circuit 52 operates. 54 operates to perform identification calculations such as similarity calculation with the 56 standard pattern and sends the results back to the controller 56.

The tying generation circuit 55 receives manual input from the controller.
It receives a format control signal and a start signal for printed Chinese characters, and outputs a timing signal to the % seed shift register and register of the sampling circuit 52.

[Other embodiments of the invention]

In the above-mentioned example, an OC in which handwritten kanji and printed kanji are mixed
Although R is taken as an example, it can be applied to OCR exclusively for printed kanji.

In the example, we explained the case of printed kanji by converting it into 2-bit multi-level data, but it is also possible to store it using even more bits, and it is of course possible that normalization and sampling accuracy will be further improved by doing so. . (If cost is ignored) This idea can also be implemented in the same way, such as outputting the input character image at the time of recognition rejection to CAT as multivalued information, or inputting high-quality images using a low group @ degree sensor. .

[Brief explanation of the drawing]

Figure @1 is a diagram illustrating handwritten and printed Kanji OCR of the prior art. FIG. 2 is a diagram illustrating an outline of an embodiment of the present invention. FIG. 3 is a diagram illustrating the cutoff and normalization section in the embodiment of the present invention. FIG. 4 is a diagram illustrating a sampling circuit in an embodiment of the present invention. FIG. 5 is a diagram explaining the functions of peripheral circuits of the sampling circuit shown in FIG. 30...Analog video signal, 31...Analog/
Digital converter, 32...Multi-value image buffer, 3
3... Cutoff control unit, 34... Printed kanji normalization pattern buffer, 35... Handwritten kanji normalization pattern buffer, 36... Normalization circuit, 37... Handwritten, printed city line . Representative Patent Attorney Noriyuki Chika (and 1 other person) Figure 1 Figure 2 Figure 3 Figure 4 Figures 4 and 5

Claims (1)

[Claims] In a character recognition device that optically scans a document and reads characters, a means for digitizing an electrical signal obtained by scanning into multi-values, a means for storing the digital signal, and a means for continuously outputting data from the raising means, It has a means for converting into a binary signal, detecting characters, determining a cutout area, and a means for adding weight of an arbitrary area for the cutout area.