JPS59200380A - Optical character reader - Google Patents

Abstract

PURPOSE:To relax a character writing condition by displaying a rejected character pattern and correcting properly and automatically the sensitivity level of a character pattern which is quantized by a correct answer fed by an operator. CONSTITUTION:An operator observes the displayed rejection patterns (b) and (c) and a form and feeds the correct answer through an answer input device 5. A correct answer pattern is sent to a level correcting circuit 6 as well as an answer editing circuit 7. The circuit 6 compares the correct answer pattern of the rejection pattern with the result of discrimination obtained from a character recognizing part 3 and decides whether the sensitivity level of a quantization pattern is proper or not. The result of this decision is fed back to a multi-value memory 2 through a signal line l3 to correct the sensitivity to a proper level. At the same time, the pattern having a sensitivity level corrected by the identification of the next character is sent to a character recognizing part 3.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an optical four-character reader (hereinafter referred to as OCR).

[Technical background of the invention and its problems]

Generally, as the scope of application of OCR expands, the quality of input data (data on forms) deteriorates. This is considered to be due to the increase in the number of OCR operators who are inexperienced with the operation. On the other hand, in order to suppress the increase in OCR costs, it is necessary to set various conditions for input data. Point 5 is that the number of characters entered on a form varies greatly depending on the person writing it and the state of the printer, and is distributed over a considerable range, so automatically correcting this requires a complicated device configuration. This is because the cost increases as a result. Therefore, in order to make the device configuration as simple as possible, it is necessary to impose conditions such as regulating the PO2 value at the entry stage. PO2 refers to print density, and is a value evaluated by (B-P)/B, where B is the background density of the form and P is the density of the print (or written characters).

Therefore, when OCR with such entry conditions set is used for various counter operations, there is no guarantee that the entry conditions will be met because the entries on forms are often made by ordinary people, and therefore reading accuracy becomes a problem. becomes.

[Purpose of the invention]

Therefore, an object of the present invention is to provide an OCR that relaxes character entry conditions without reducing reading accuracy.

[Summary of the invention]

In order to achieve the above object, the OCR according to the present invention displays a rejected character pattern on a display device, an operator inputs a correct answer based on the displayed content, and the quantized character pattern is quantized by the input correct answer. The present invention is characterized in that it is configured to appropriately automatically correct the sensitivity level of the character pattern.

[Embodiments of the invention]

FIG. 1 shows an embodiment of OCR according to the present invention.

The read data D is converted into a multilevel quantization pattern by the photoelectric conversion section 1 and then stored in the multilevel memory 2. The stored patterns are sequentially cut out and sent to the character recognition unit 3 via a signal line! Sent via 1.

The character recognition unit 3 performs discrimination by comparing the extracted pattern with an internal dictionary. From now on, set the first small letter to a high level, and reject anything you are unsure about.

The rejected pattern, which is used to correct the sensitivity level to be described later, is notified to the multi-level memory 2 through the signal edge j2 and displayed on the pattern display bag g (CRT display device or the like) 4.

The pattern display device 4 displays a reject pattern as shown in FIG. That is, FIG. 2(a) shows manuscript characters written on a book cover, with the number "8" having a faded part A and a smeared part B.

Assume that C exists. A nokuturn obtained by quantizing the original characters at the first sensitivity level is shown in FIG. 2(b), and a nokuturn obtained by quantizing the original characters at the second sensitivity level is shown in FIG. 2fc). The first sensitivity level is darker than the second sensitivity level. The above fa) (b) Nokuturn display device 4 quantized at two sensitivity levels shown in the figure.
will be displayed. The reason why the turn is displayed in two sensitivity levels, dark and dark, is to make it easier and more accurate for the operator to make a visual judgment.

The operator inputs the correct answer using the answer input device 5 by looking at the displayed turn (Fig. 2 (b), (c)) and the form.The answer input device 5 may be a keyboard, voice input device, etc. (X, ) can be done. The correct answer, that is, the correct answer, is sent to the level correction circuit 6 and the answer editing circuit 7.

The level correction circuit 6 compares the correct pattern for the relevant redirection turn with the discrimination result of the character recognition section 3, and determines whether the sensitivity level of the quantization pattern is appropriate. The result of that judgment is the signal line! 3 to the multi-value memory 2, and the sensitivity level is appropriately corrected based on the feedback judgment result. After the next character is identified, the corrected sensitivity level pattern is sent to the character recognition unit 3. The reason for uniformly identifying subsequent character patterns using the corrected sensitivity level is that the density level of characters written by the same person is generally constant within the same document7. This is because the same applies to the output (printing) of . Even if the application is rejected, there is no problem because the same correction as described above is automatically performed.

The answer editing circuit 7 selects the correct answer from the answer input device 5, and
Character identification section 3 Finally edits the answer based on the identification number turn and outputs the same stone.

〔Effect of the invention〕

As described above, according to the present invention, since the sensitivity level is corrected based on the redirect turn and taking into account the recognition ability of the operator, highly accurate identification is possible. As a result, accurate recognition can be achieved without any restrictions such as particularly strict conditions when filling out forms.

Further, since automatic correction is performed in this manner, patterns with a wide range of density can be recognized without adding expensive circuits.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the main part of the OCR according to the present invention, and FIG. 2 is an explanatory diagram of the beam target pattern, in which (a) is a document character pattern, (b) is a display pattern of the first sensitivity level, and (C ) are diagrams showing display patterns of the second sensitivity level, respectively. 1...Photoelectric conversion unit 2...Multi-level memory 3...Character recognition unit 4...Pattern display device 5...Answer input device 7...Answer editing circuit Applicant's agent Kiyoshi Inomata

Claims (1)

[Claims] A photoelectric conversion unit that photoelectrically converts read data and quantizes it into a multilevel pattern; a multilevel memory that stores the quantized pattern; and a photoelectric conversion unit that cuts out a stored character pattern from the multilevel memory. a display device for displaying the rejected noku turns in at least two density levels, and a display device for inputting a correct answer for the rejected noku turns according to an operator's judgment; A re-input device compares the correct answer for the reject pattern with the recognition result in the character recognition unit, determines whether the sensitivity level of the quantization number stored in the multi-level memory is appropriate, and makes the determination. Level correction that corrects the sensitivity level based on the result - Correct answer and recognition of the circuit and reject pattern