JPH0136154B2 - - Google Patents

Description

[Detailed Description of the Invention] The present invention relates to a character recognition method, and more specifically, to a character recognition method that is optimal for an optical character reading device (hereinafter referred to as OCR) that can read handwritten characters. .

In OCR, in order to reduce the capacity of the recognition dictionary and shorten the recognition time, normalization that adjusts the size and slope of character patterns has become established as essential.

In particular, in the case of OCR, which reads handwritten characters, normalization is extremely important because with the recent spread of OCR, it may be difficult to provide sufficient training to people filling out forms.

However, although it is relatively easy to normalize the size of a character, it is difficult to normalize the slant of a character at the preprocessing stage when it is unclear what the character is called, and this can lead to misreading. There is a problem with becoming.

In view of these problems, the present invention was made by focusing on the fact that even if the characters are different, the degree of inclination of handwritten characters remains constant depending on the habits of the person filling out the form and the habits of the group or organization to which they belong. It is.

The first object of the present invention is to make the normalization accurate by performing normalization by taking into account the slope of the reference character written on the form, and to accurately and quickly recognize characters with a small capacity recognition dictionary. It is about making recognition possible.

A second object of the present invention is to enable quick character recognition by limiting the use area of the recognition dictionary by taking into account the slope of the reference characters written on the form. .

Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention, in which 1 is a photoelectric conversion section, 2 is a character cutting section, 3 is a normalization section, 4 is a recognition section, 5 is a dictionary memory, and 6 is a tilting section. 7 indicates a register.

The photoelectric conversion unit 1 is a circuit that converts reflected light from a form into an electrical signal and converts the form image into a quantum pattern.

The character cutting section 2 is a circuit that cuts out the quantum pattern characters one by one, and the cut out character patterns are sent to the normalization section 3 and the slope detection circuit 6.
It is designed to be supplied to

The normalizing section 3 adjusts the size and inclination of the character pattern supplied from the character cutting section 2, and the normalized character pattern is supplied to the recognition section 4.

The recognition unit 4 performs pattern recognition of the normalized character pattern by matching the normalized character pattern with a standard pattern stored in the recognition dictionary 5. The recognition unit 4 is configured to supply a control signal to the slope detection circuit 6 when it obtains an answer of "1" as a result of recognition.

The recognition dictionary 5 is a memory in which standard patterns are stored.

The slope detection circuit 6 is a circuit that, when supplied with a control signal from the recognition section 4, calculates the slope of the character pattern "1" supplied from the character cutting section. The obtained degree of inclination is expressed, for example, as an 8-way direction code or a 16-way direction code.

The register 7 is for storing information indicating the slope supplied from the slope detection circuit 6, and the information indicating the slope stored in the register 7 is supplied to the normalization unit 3 and the recognition unit 4. There is.

Next, the operation of this embodiment will be explained.

When the OCR starts operating, the character pattern quantized by the photoelectric conversion unit 1 is extracted character by character by the character extraction circuit 2, and the normalization unit 3
supplied to

The character pattern supplied to the normalization section 3 is normalized and then supplied to the recognition section 4, where it is recognized. This has been known for a long time
This is similar to the basic operation of OCR, but in this embodiment, when the recognition unit 4 obtains an answer of "1" as a result of character recognition, the recognition unit 5 sends the slope detection circuit 6
A control signal is supplied to the

On the other hand, the inclination detection circuit 6 includes a character cutting section 3.
When the character pattern is supplied from the recognition section 4, and the slope detection circuit 6 is supplied with the control signal from the recognition section 4,
The degree of inclination of the character pattern "1" supplied from the character cutting section 3 is determined, and information indicating the degree of inclination is supplied to the register 7.

The information indicating the degree of inclination supplied to the register 7 in this manner is held in the register 7 for a predetermined period of time.

Note that the predetermined period here may be a period for reading one form or a period for reading one job of forms.

The information indicating the slope supplied to the register 7 is
Thereafter, each time the normalization unit 3 normalizes a character pattern and each time the recognition unit 4 recognizes a character pattern, the information is supplied to the normalization unit 3 and the recognition unit 4.

Thereafter, the normalization unit 3 normalizes the character pattern by taking into account the information indicating the slope. For example, if the information indicating the degree of inclination indicates that the character pattern "1" is slanted to the right, the normalization unit 3 determines that the characters of the person filling out the form tend to be slanted to the right. Normalize the character pattern as a standard.

On the other hand, from now on, the recognition unit 4 also takes into consideration the information indicating the degree of inclination, corrects the degree of inclination, and performs pattern recognition of the character pattern.

Next, another embodiment will be described.

FIG. 2 is a block diagram showing another embodiment of the invention.

Components that are the same as those in the first embodiment are numbered the same as in FIG. 1 to avoid duplication of explanation.
53 is a recognition dictionary, 8 is a switching circuit, 9 is a format control circuit, and 10 is a multiplexer.

The recognition dictionaries 51, 52, and 53 are memories in which standard patterns for character recognition are stored. Among these, the recognition dictionary 51 is a dictionary that assumes a person who fills out forms whose characters tend to lean to the left.
Reference numeral 2 is a dictionary intended for a person who fills out a form whose characters tend not to be particularly inclined, and the recognition dictionary 53 is a dictionary which is assumed to be a person who fills out a form whose letters tend to be inclined to the right.

The switching circuit 8 is a circuit that determines whether the information indicating the slope of the character pattern detected by the slope detection circuit 6 is used for normalization or for selecting a recognition dictionary.

The format control circuit 9 is used to control the format and the decision made by the switching circuit 8 when cutting out characters.

The multiplexer 10 has recognition dictionaries 51, 52,
This circuit performs 53 selections.

Further, FIG. 3 shows an example of the form 11 used in the embodiment shown in FIG.
3.14 indicates an information field.

Next, the operation of this second embodiment will be described.

First, the person filling out the form enters "1" as a reference character in the reference character entry frame of the form 11, and then enters necessary information in the information fields 13 and 14. Third
As shown in the figure, as long as the person filling out the form is constant, the slope of the standard character written in the standard character entry frame 12 and the slope of the characters written in the information fields 13 and 14 will almost match. This will be generally understood without any special explanation. In the example of FIG. 3, the characters of this form filler tend to slant to the right.

The image of the form 11 filled in in this way is patterned by the photoelectric conversion section 11 and supplied to the character cutting section 2.

At this time, format information is supplied to the character cutting section 2 from the format control section 9.
The character cutting section 2 cuts out the character pattern "1" written in the reference character entry frame according to the format information, and the cut out character pattern is supplied to the slope detection circuit 6.

The slope detection circuit 6 determines the slope of the character pattern "1" and holds information indicating the calculated slope.

Next, the format control section 9 supplies a control signal to the switching circuit 8, instructing whether to supply the information indicating the slope to the normalization section 3 or the multiplexer 10. Note that this control signal is based on an operator's instruction.

First, a case will be described in which the control signal supplied from the format control section 9 instructs the normalization section 3 to supply information indicating the degree of inclination.

Every time the normalization unit 3 normalizes a character pattern, information indicating the slope is supplied from the slope detection circuit 6 to the normalization unit 3 via the switching circuit 8.

Thereafter, the normalizing section 3 normalizes the character pattern supplied from the character cutting section 2, taking into consideration the information indicating the slope.

Next, the control signal supplied from the format control section 9 transmits information indicating the slope to the multiplexer 10.
We will explain the case of instructing what should be supplied to.

Every time a character is recognized by the recognition unit 4, information indicating the degree of inclination is supplied from the degree of inclination detection circuit to the multiplexer 10 via the switching circuit 8.

Based on this information, multiplexer 10 selects a recognition dictionary. When reading the form shown in FIG. 3, the recognition dictionary 53 is selected.

On the other hand, the normalized character pattern is supplied to the recognition unit 4 from the normalization unit 3, and the recognition unit 4 matches the supplied character pattern with the character pattern stored in the recognition dictionary 53.

As explained above, according to the present invention, character patterns can be normalized by taking into account the slope of the reference character, so accurate normalization is possible without complicated procedures, and Dictionaries enable accurate and quick character recognition.

Furthermore, according to the present invention, the area in which the recognition dictionary is used can be limited by taking into consideration the slope of the reference character.
The number of matching operations is reduced, making character recognition more efficient.

In addition, although two examples were shown above,
Various other combinations of applications are possible.

[Brief explanation of drawings]

1 and 2 are block diagrams showing embodiments of the present invention. FIG. 3 is an explanatory diagram of a form. DESCRIPTION OF SYMBOLS 1... Photoelectric conversion part, 2... Character cutting part, 3... Normalization part, 4... Recognition part, 5, 51, 52, 53... Dictionary for recognition, 6... Slope detection circuit, 7... Register,
8...Switching circuit, 9...Format control section, 1
0...Multiplexer.

Claims (1)

[Scope of Claims] 1. A character recognition method characterized by detecting the degree of inclination of a reference character on a form and performing character recognition by taking the detected degree of inclination into account. 2. A character recognition method according to claim 1, characterized in that characters are normalized and recognized by taking into account the detected degree of inclination. 3. A character recognition method according to claim 1, characterized in that a recognition dictionary is selected in consideration of the detected degree of inclination and character recognition is performed. 4. In the character recognition method described in Claims 1, 2, or 3, a reference character entry frame is provided on a form, and the character entered in the reference character entry frame is used as a reference character to determine the degree of inclination. A character recognition method that detects.