JPS5894064A - Handwritten character recognizer - Google Patents

Abstract

PURPOSE:To recognize the handwritten characters with high speed and stability, by reading previously the character samples of a writer to produce an exclusive dictionary of the writer and therefore using the dictionary to recognize the handwritten characters of the writer. CONSTITUTION:The operation of a key is detected at a keyboard part 14, and the register control and register number signals are fed to a control part 9. The character samples of a writer are read through a photoelectric converting part 1, and the features of the writer's sample pattern are discriminated through a feature matrix extracting part 7, a discriminating part 8, etc. Then the writer's exclusive dictionary is stored in a memory part 10 through a control part 9. The contents of the dictionary are read out to the part 9 when the operation of a key is detected through the part 14. The characters on a slip, etc. which are read through the part 1 are processed through a character frame detecting part 4, subpattern extracting part 6, matrix extracting part 7, stroke extracting part 12, average angle calculating part 13, etc. to be applied to the part 9. The writer's exclusive dictionary sent from the memory 10 is used at the part 9 to recognize the handwritten characters in a quick and stable way.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a high-speed and highly accurate handwritten character recognition device.

In conventional handwritten character recognition devices, in order to absorb variations in features due to variations in the slope of character lines due to different writers, the variations in features are absorbed by using a plurality of dictionary masks. However, this device has the disadvantage that the time required to compare the extracted features with the dictionary during identification increases in proportion to the number of dictionary masks, resulting in a reduction in the processing speed of the device.

In order to eliminate such conventional drawbacks, the present invention creates a dictionary exclusively for the writer by reading the writer's character sample in advance, and uses the dictionary to read the handwritten characters of the writer. The purpose is to provide a high-speed and stable handwritten character recognition device.

FIG. 1 is a block diagram showing one embodiment of the handwritten character recognition device of the present invention, in which 1 is a photoelectric conversion section, 2 is a pattern register, 3 is a line width calculation section, 4 is a character frame detection section, and 5 is a Character frame division determination unit, 6 is a subno or turn extraction unit, 7 is a feature matrix extraction unit, 8 is an identification unit, 9 is a control unit, 10 is a memory unit, 1 is a character name output, 12 is a stroke extraction unit,
Reference numeral 13 indicates an average angle calculation section, and 14 indicates a key date section.

Next, its operation will be explained. First, (1) the author's own dictionary creation, and then (2) the recognition operation.

To create a dictionary exclusively for the author (1), follow the steps below.

The operator presses the writer registration control key and the writer registration number key on the keyboard section J4. key? - code section 14
detects the depression of the key and sends a registration control signal and a registration number signal to the control 9. Next, the operator sets the form containing the writer's character sample into the reading mechanism.

In this example, the character sample is the kanji character ``木''. The characters on the form set in the reading mechanism are converted into binary quantized digital electrical signals by the photoelectric conversion section 1 and stored in the pattern register 2. At the same time, the line width of the input pattern is calculated in the line width calculation section 3. The sub-pattern extraction unit 6 performs vertical scanning of the pattern register all at the same time, and extracts a vertical sub-pattern (vsp) based on the relationship between the continuous length of black bits and the line width calculated by the calculation unit 3. Similarly,
horizontal scan/horizontal turn (H4F),
A right diagonal subpattern (H8
F), a left diagonal sub-pattern (LSP) is extracted by performing a left diagonal 45° scan. Figure 2 is an example of the original pattern and each supno and turn. (,) is the original/mother turn, (b) is the vertical sub-pattern (vsp), and (C) is the horizontal sub-Al-:/
(H4F), (d) is right diagonal sub/, o turn (H
8F) , (e) C is a left diagonal Mesa No J turn (LSP). The stroke extraction part 2 12 performs horizontal or vertical scanning for each sub-pattern register, detects the change point from a white point to a black point (the character line part is the black point), and detects the change point from a black point to a white point, and Row) Strokes are extracted from the relationship between the number of change points and change point coordinates in the previous scan and the number of change points and change point coordinates in the current column (or row), and the pattern of both end points of the stroke in each extracted subpattern register. The average angle calculation unit 1 calculates the coordinates in the two-dimensional coordinate system defined by the register (the origin is at the lower left of the pattern register).
Send to 3. The average angle calculation unit 13 refers to the coordinates of both end points of the stroke in each subpattern register extracted by the stroke extraction unit 12, and calculates the average angle for each subpattern. First, the coordinates of both end points of the stroke extracted from the horizontal sub-pattern are (HXSn, HYSn), (
HXEn, HYEn) However, n = 1,...N, N
When is the number of strokes, the average angle θ□ is calculated using equation (1). (However) (XEn>HXSn) 09. θR1θ is calculated using equations (2) to (4).

(However, VYEm) VYSm, RXEI) RXS
I, LXEk)LXSk) However, in the above formula, M and L are vertical sub-patterns, right diagonal sub-turns,
This is the number of strokes extracted from the left diagonal sub/g turn. The average angle calculation unit 13 sends the average angle of each sub-pattern calculated using the above formula to the control unit 9. The control section 9 refers to the average angle of each sub-pattern sent out from the average angle calculation section J3 and the aforementioned registration control signal and registration number signal sent out from the key gate section 14 to create an author-specific dictionary.

In this example, there are three types of dictionaries (hereinafter referred to as original dictionaries) that are the basis for creating the author's own dictionary: horizontal feature matrix mask, vertical feature matrix mask, right diagonal feature matrix mask, and left diagonal feature matrix mask. A mask corresponding to the average angle of is prepared. However, the horizontal feature matrix is a feature matrix extracted from horizontal subpatterns, the vertical features are extracted from vertical subpatterns, the matrix is extracted from vertical sub-turns, the right diagonal feature matrix is extracted from right diagonal subpatterns, and the left diagonal feature matrix is extracted from left diagonal subpatterns. The extraction method will be described later.

The average angle of each of the three types of feature matrix masks is θ□ -o25. −o25<θ□<0.25,
Three types of θv -0,25, -0,25, 0.25<θ□
<θ, (0, 25, 0, 25<, θ9 three types θ, <
0.7, o7<θ, <1.4, 1.
4 <08 three types θ, <-1, 4, -t4<θt, <
(We prepared mars and ri corresponding to three types: 17, 0, 7<, and θ1.

The control unit 9 reads out the original dictionary from the memory unit io and calculates θ□ for the scribe calculated by the average angle calculation unit 13.
, θ7. θ8. θH1θ9. of the original dictionary for each of the IJ mask with reference to θ1. θ8
．． With reference to the classification of θ1, the corresponding mask is extracted from the original dictionary and configured as a set of dictionaries, and the key is stored at a predetermined address in the memory section No. 0 corresponding to the registration number sent from the keypad section 14. do.

Next, the recognition operation (2) is performed by the following operation.

The oiler presses the document author registration number key and the read control key of the key gate unit 14 to set the document to be read in the reading mechanism. When the keyboard unit 14 detects the pressing of the key and receives the read control key pressing signal, the registration number signal, and the registration number signal, it reads out the dictionary from the memory unit 10 at a predetermined address corresponding to the registration number, and reads the dictionary from the memory unit 10 at a predetermined address corresponding to the registration number.
After sending the dictionary to the identification unit 8, the identification unit 8 waits for the category name to be output.

The characters on the form set in the reading mechanism are converted into binary quantized digital electrical signals by the photoelectric conversion section 1 and stored in the pattern register 2. At the same time,
A line width calculating section 3 calculates the line width of the input pattern. Subno R turn extraction section 6Vi performs a vertical scan on the A' turn resist and extracts a vertical subpattern (vsp) based on the relationship between the length of the black bit mountain range and the line width calculated in calculation section 3. . Similarly, the horizontal sub-pattern (LSP) is obtained by horizontal scanning, and the right-diagonal sub-pattern (LSP) is obtained by scanning 45 degrees diagonally to the right.
Wow, the left diagonal sub-pattern (
LSP).

The character frame detection section 4 detects a character frame circumscribing the character pattern in the pattern register 2, and sends the result to the character frame division determination section 5. The character frame division determining unit 5 divides the detected character frame into an M×N area (M and N are integers; in this embodiment, M=N=
5) Determine the dividing point coordinates on the X and Y axes for dividing. Here, the X axis indicates the horizontal direction of the character frame, and the Y axis indicates the vertical direction.

The feature matrix extraction unit 7 generates fi VSP and LSP based on the division point coordinates determined by the character frame division determination unit 5.

Divide the character frame area on each subnof turn register of LSP and LSP into MxN areas, count the number of black bits Bij in each area, and use the line width W to indicate the character line length by equation (5). The features are calculated and a MxNx4-dimensional feature matrix is created.

Lij= JJ/w (5) After that, the VSP feature matrix is the length ΔY of the character frame in the Y-axis direction, the H8P feature matrix is the length ΔX in the X-axis direction, and the LSP and LSP % feature matrix are ( Δχ+ΔY)
Normalize each with /2 and finally M
4-dimensional feature map) Create IJ Kusu. The identification unit 8 uses the dictionary mask (f,) and the extracted feature matrix C.
The distance (D) defined by equation (6) is applied between the distance (D) and f1), and the category name of the dictionary mask for which D is the minimum value is sent to the control unit 9. The control unit 9 is configured to output the received category name to a character name output.

D-〆ΣCf, -f2(6) In this way, by reading the scribe's character samples, the tendency of the slope of the writer's character lines is extracted, and a dictionary dedicated to the writer is created using multiple pre-prepared dictionaries. Since the scribe's characters are created by selecting from a dictionary/calligraphy mask and the dictionary is used to recognize the scribe's characters, unnecessary dictionary masks are removed to recognize the scribe's characters. , can perform high-speed processing. In addition, since a dedicated dictionary for each scribe can be created and registered, it is possible to use the most suitable dictionary for each scribe for recognition. .

As explained in detail above, the present invention creates a dictionary exclusively for scribes and uses this dictionary to recognize characters.
It is possible to perform high-speed and highly accurate recognition, and is highly effective when used in a high-speed and highly accurate handwritten character recognition device.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the handwritten character recognition device of the present invention, and FIG. 2 is a diagram showing an example of an original pattern and each sub-pattern. l...Photoelectric conversion unit, 2...Pattern register, 3.
... line width calculation section, 4 ... character frame detection section, 5 ... character frame division determination section, 6 ... subpattern extraction section, -7 ...
・Feature matrix extraction unit, 8...Identification unit, 9...Control unit, 1o...Memory unit, 11...Character name output, 1
2... Stroke extraction section, 13... Average angle calculation section, 14... Key is - mode section.

Claims (1)

[Claims] Several types of dictionary masks are prepared for each feature, which are classified and created in advance for each feature.The author's character samples are photoelectrically converted and quantized, and the resulting digital signal is stored as an original pattern in a pattern register. The stroke components in each direction are extracted from the original pattern, the slope of the character line in each direction is extracted, the classification for each feature for the writer is determined, and a dictionary dedicated to the writer is created using the plurality of characters for each feature prepared in advance. Select each feature from the dictionary mask,
A handwritten character recognition device characterized in that the device is configured and registered and recognizes characters using the dictionary.