JPS58106666A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To prevent the recognition rate from being decreased remarkably after the recognition of a handwriting written in a very peculiar way, by discriminating and controlling the correction or addition of a characteristic dictionary when a pattern impossible for recognition is produced, from the operator. CONSTITUTION:A characteristic parameter of a character pattern impossible for recognition is read out from a storage device B 108, this characteristic parameter is converted into an alphanumeric code at a conversion circuit 109 via a correction control circuit 107 and given to an input/output processing circuit 10. Thus, a resoration pattern of the character pattern impossible for recognition and characteristics of each stroke are displayed on a display 111. The operator discriminates the character types of the character pattern from the display on the display 111 and inputs the character code from a keyboard 112. The circuit 107 reads out the standard characteristic parameter corresponding to the code from a storage device 105 and displays it on the display 111 via the circuit 109. Then, the operator discriminates whether the characteristic parameter after correction is to be additionally registrated in the device A or to be revised, and performs the additional or revised registration through the device 112.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention 14 recognizes patterns such as characters. The present invention relates to an S-turn recognition device, and particularly to a pattern recognition device that performs pattern recognition by extracting feature parameters from a pattern to be recognized and comparing them with feature parameters registered in a feature dictionary.

The general configuration of such a pattern recognition device will be explained with reference to FIG. 1, taking an optical character recognition device as an example.

Ilv Ihachi) to I station 2 几子 1jJV'-Kawamoshi-(
It is a scanner that reads the image of ΔKatana 1 Nemikoredo. The image information read by this is binarized by the preprocessing circuit 2,
It is subjected to character extraction and other preprocessing, and is sent to the feature extraction circuit 3 as character pattern information for each character.

The feature extraction circuit 3 extracts feature parameters from the input character pattern and sends them to the matching circuit 4. There is still no storage device (
Standard feature parameters for each character registered in advance are read out from the feature dictionary 5 and sent to the matching circuit 4.

The matching circuit 4 compares the extracted feature parameters with the feature parameters given from the storage device 5, and if matching is achieved with the standard feature parameters of a certain character, the matching circuit 4 recognizes the character pattern currently being recognized as that character. , the character code is transferred to an external storage device or the like through the output circuit 6. If matching cannot be achieved with the standard feature parameters of any character, it is determined that it is impossible to determine.
The corresponding read character pattern is sent to the display device 8 via the display control circuit 7 and displayed, and character recognition is left to the operator.

The operator deciphers the character pattern displayed on the display device,
The character code is input from the keyboard device 9. The input character code is sent to the output circuit 6.N through the display correction circuit 7 and transferred to an external storage device or the like. This completes the correction process for one character that cannot be recognized by the recognition device.

In addition, there is a method in which the operation of the device is interrupted every time an unrecognized character pattern occurs, and a method in which even if an unrecognized character pattern occurs, it is skipped and the recognition operation continues from the next character pattern. After completing the processing of a batch (one or more documents), a batch processing device requires a relatively large-capacity image storage device to store unrecognizable character patterns.

Devices with learning functions are still being devised to improve recognition rates. This calculates the distance between the feature parameter extracted from the unrecognizable pattern and the standard feature parameter registered in the feature dictionary, and at the same time recognizes the pattern as corresponding to the standard feature parameter with the minimum distance. Correct the corresponding standard feature parameter so that it is smaller and re-register it in the feature dictionary.

However, when the contents of the feature dictionary are automatically changed using such a learning function, another problem arises. That is,
After recognizing the handwritten characters of a person with a pronounced habit of writing, the feature parameters in the feature dictionary deviate considerably from the standard. Unfortunately, if you subsequently try to recognize another person's handwritten characters, the recognition rate will drop significantly even if that person's font is standard. Therefore, there is a practical problem in allowing the contents of the feature dictionary to be modified unconditionally.

In view of the above points, the present invention allows the operation m to modify or add the contents of the feature dictionary when an unrecognizable turn occurs.
The present invention aims to provide an S-turn recognition device that can be easily judged and controlled individually, and can also perform batch processing without requiring a large-capacity image storage device.

The present invention (the pattern recognition device based on r) extracts feature parameters from the pattern to be recognized and compares them with the ζ parameters (features stored in the feature dictionary).
The basic structure of recognizing the recognition target nositane is the same as the conventional one, but the feature that is not found in the conventional one is that it includes the following components. That is, there is a means for restoring the feature parameter extracted from the recognition target pattern that could not be recognized by comparing the % signature parameters into an S turn and displaying it on a display means, and a means for displaying the feature parameter ζ parameter currently being displayed on the display means. , a means for modifying according to a command input by an operator through the input means, and a means for modifying the feature parameter Z parameter modified by this means into a feature dictionary according to a command input by the operator through the input means. and a means for storing (adding or changing) the information.

Hereinafter, the present invention will be described in detail with respect to one embodiment.

FIG. 2 is a block diagram of an optical character recognition device according to the present invention.

toi is a scanner that optically scans and reads input forms. The image information read by the scano 101 is sent to a preprocessing circuit 102 and undergoes preprocessing such as binarization and character extraction. Character pattern information output character by character from the preprocessing circuit 102 is sent to a feature extraction circuit 103, where feature parameters are extracted. In this example, the strokes on the azimuth side that make up the character pattern, the center of gravity of each stroke,
The length is extracted as a feature parameter.

The extracted feature parameters are sent to the matching circuit 104 and sequentially compared with the feature parameters read from the storage device A (feature dictionary) 105. When a certain character is matched with the characteristic parameters registered in the storage device A 105, the character code of that character is output to the output circuit 106.
The data is transferred to an external storage device, etc. In the case of a character pattern that cannot be matched and cannot be recognized, the feature parameters extracted from the character pattern are sent from the matching circuit 104 to the storage device B via the correction control circuit 107.
108.

The storage format of the characteristic parameters in the storage device B 108 and the storage device A 105 is, for example, as shown in FIG. The length is data compressed and stored to increase storage efficiency. Such data-compressed characteristic parameters are difficult for humans to read even if they are displayed as they are on the display device 111. However, in the storage device 8108, the character code is a modified code until input by the operator.

In addition, when an unrecognizable character pattern occurs, there is a sequential mode that interrupts the recognition operation and requests correction by the operator, and a sequential mode that terminates the recognition operation for one batch (input form of one page or multiple pages). There are a batch mode that requests correction by the operator, and which mode to select can be specified from the keyboard device 11.2 to the correction control circuit i07. When the sequential mode is specified, the correction control circuit 10
7 immediately interrupts the recognition operation when an unrecognizable character pattern occurs, and starts a correction operation to be described later. Resuming the recognition operation is made to wait until an instruction is given from the keyboard device 112. If the batch mode is designated, the correction control circuit 107 starts the correction operation when recognition operations for one batch are completed.

In the correction operation, the feature parameters of the unrecognizable character pattern are read from the storage device 81.08, and the conversion circuit 109 expands the feature parameters into a stroke dot pattern via the correction control circuit 107.
Display device 1 displays the number, direction, center of gravity, and length of each stroke.
11 is converted into an alphanumeric code that can be displayed, and sent to the input/output processing circuit IIO. As a result, the unrecognizable character pattern 1 Natsumoto pattern and the characteristics of each stroke are displayed on the display device 111. The operator determines the character type of the unrecognizable character pattern from the restored pattern displayed on the display device 111, and inputs the character code from the keyboard device 112. The input character code is sent to the correction control circuit 107 via the input/output processing circuit 110 and the conversion circuit 109. Modification control circuit 107 reads standard characteristic parameters corresponding to the character code of the letter from storage device AlO3 and sends them to conversion circuit 109. The conversion circuit 109 converts the manually entered standard feature parameters into alphanumeric codes, sends them to the input/output processing circuit 110, and displays them on the display device 1. An example of the display contents at this stage is shown in FIG.

The example shown in FIG. 4 is a case where the pattern of the character "A", which has a strong character as shown in FIG. 5 (a), becomes unrecognizable. The feature extraction circuit 103 extracts strokes in four directions from DI to D4 as shown in FIG. 6 from the character pattern, so from the character pattern in FIG.
The stroke features of are extracted.

Therefore, the feature parameter of this character pattern is the fourth
It will be displayed as shown in the bottom part of the figure. Here, NN is the number of strokes, 5l-84 is the number of each stroke ■~■, the first number within 0 following the stroke number is the center of gravity ° coordinate of the stroke, and the first number within next 0 is the stroke The next number is the stroke length. The upper left pattern in FIG. 4 is a pattern restored from the characteristic parameters of the unrecognizable character pattern. In addition, in practical terms
d. The handwritten character pattern has variations in the thickness of the character lines,
Noise components such as bending and bouncing are often present, and some of them may be extracted as strokes. In this case, the unnecessary strokes will also be included in the restoration pattern.

FIG. 5(B) shows a standard pattern for the letter "A", which consists of three strokes - - -.

A feature parameter for the pattern of the standard character "A" is registered in advance in the storage device AlO3, which is a feature dictionary, as a standard feature parameter. This standard feature parameter is displayed as shown in the upper right corner of FIG.

Since it is displayed as described above, the operator can easily know the correspondence between the unrecognizable character pattern and its characteristic parameters, and can also easily understand the difference from the standard characteristic parameters. The operation determines whether there is anything that should be considered noise in the feature parameters extracted from the unrecognizable character pattern, and if there is noise, the input/output processing circuit 11 is operated from the keyboard device 112 to delete it.
Input the command to 0. When this command is given, the input/output processing circuit 110 modifies the characteristic parameters on the internal buffer register according to the command, and displays the modification results on the display device 111.
to be displayed. When the necessary corrections are completed, the operator determines whether the corrected characteristic parameters should be additionally registered in the storage device A 105, whether the characteristic parameters should be updated and registered as standard characteristic parameters, or whether registration is unnecessary. Commands are input through device 112.

When an additional registration command is input, input/output processing circuit 1.
The modified feature parameters (including character codes) in the 10 buffer registers are sent to the conversion circuit 109;
After being reversely converted to the storage format in the storage devices A and B, it is additionally stored in the storage device AlO3 via the modification control circuit 107, and the modification code in the corresponding area of the storage device 8108 is rewritten to a character code. When an update registration command is input, the area of the corresponding character code in the storage device AlO3 is rewritten using the above-mentioned inversely converted feature parameters.

The same applies to storage device B 108. When a command that does not require registration is input, only rewriting of the storage device B108 is performed.

When this operation is completed, the modification control circuit 107 checks whether there is a modification code in the next area in the storage device B 108, and if there is a modification code, reads out the characteristic parameters following it and sends it to the conversion circuit 109, and performs the modification as described above. Repeat the action. When there are no correction codes in the storage device Bi12, the correction control circuit 107 sequentially reads character codes (recognition results by the operator) for unrecognizable character patterns from the storage device B 108 and transfers them to the outside via the output circuit 106.

In the sequential mode, when any of the above commands is input, the character code is immediately transferred to the outside after performing the necessary operation on the storage device AlO3, and the interrupted recognition operation is restarted.

As described in detail above, in the turn recognition device according to the present invention, since the operator can check and control changes in the contents of the feature dictionary (105), it is possible to check and control the change in the contents of the feature dictionary (105). This can prevent problems such as the recognition rate for other people's handwritten characters dropping significantly after recognition. However, since the pattern can be restored from the feature parameters extracted from the unrecognizable character pattern, there is no need to save the unrecognizable character turns themselves even when using a batch processing method, and data compression is possible. Since it is only necessary to store the characteristic parameters that have been determined, the required capacity of the storage device can be significantly reduced.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing the general configuration of a conventional optical character recognition device, Fig. 2 is a block diagram showing an example of an optical character recognition device according to the present invention, and Fig. 3 shows the storage format of feature parameters. Figure 4 is a diagram showing an example of the display during correction, and Figures 5 (a) and (b) show the non-standard pattern of the letter "A" and the standard 2 turns and extracted from them. FIG. 6, which is a diagram for explaining the stroke, is a diagram showing the direction of the stroke. 101...Scanner, 102...Preprocessing circuit, 10
3... Feature extraction circuit, 104... Matching circuit,
105... To storage device, 106... Output circuit, IO
2...Modification control circuit, 108...Storage device, 109
Conversion circuit, iL. ...Input/output processing circuit. Agent Patent Attorney Makoto Suzuki

Claims (1)

[Claims] 1. By comparing the feature parameters extracted from the recognition target pattern with the feature/ζ parameters stored in the storage device, the pattern recognition device that recognizes the recognition target turn can determine the above feature parameters. A means for restoring the feature/bimeter extracted from the recognition target pattern that could not be recognized in the comparison into a pattern and displaying it on the display means, and a feature parameter being displayed on the display means is inputted by the operator through the input means. means for modifying according to the command and the characteristic parameters modified by this means;
It is characterized by comprising means for storing in the storage device according to a command input by an operator through the input means. C-turn recognition device.