JPS63208183A - Character recognizing system - Google Patents

Abstract

PURPOSE:To reflect a feature of inclination even in case of character normalization where the coordinate value of a partial pattern is divided by the character width to normalize the partial pattern, by subjecting the partial pattern of an approximately vertical or horizontal line to square normalization. CONSTITUTION:For example, KANJI (Chinese character) 'in' is divided, and a partial pattern Q is calculated by character width normalization with respect to KANJI 'in'. The partial pattern Q is calculated in a square normalization part 4b and a partial pattern Q value calculating part 5a by square normalization with respect to a right-hand radical '¦'. The distance of the partial pattern Q value is calculated by matching between resultant partial pattern Q values and partial pattern Q values of registered patterns in a matching part 5b and is defined as the distance of KANJI 'in'. Thus, the partial pattern Q valve approximates that of a registered pattern even if the right-hand radical '¦'is subtly inclined in either direction.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a character recognition method, and more particularly to a character recognition method for recognizing written characters using a feature parameter such as a partial pattern Q value.

(Prior art) Conventionally, this type of character recognition device uses a pre-processing unit that removes unnecessary data from a coordinate data string obtained by handwriting input on a tablet and performs linearization processing, and a one-sided processing unit. and a feature point extraction unit that extracts feature points representing characteristics of strokes forming a written character from the linearized coordinate data string,
Characters are recognized from the feature points extracted by the feature point extraction unit using parameters called partial pattern Q values representing characteristics such as the length, direction, and position of each segment for each partial pattern.

The relationship between strokes, segments, and partial patterns in the following is as follows: segment < stroke = partial pattern. Here, a segment refers to a line segment that connects a feature point calculated by the feature point extraction unit to the next feature point, and a stroke refers to
A collection of several segments, from Ben-on to Ben-off.

Further, a partial pattern is a collection of several strokes that generally constitute edges or structures, and refers to a character that is composed of multiple partial patterns.

The partial pattern Q value is calculated as follows after the coordinate values of the partial pattern are normalized by the character width normalization method. Note that in the formula below, Σ indicates addition regarding all strokes and all segments.

Further, HX and HY indicate the character width in the X direction and the Y direction, respectively.

■+X-direction position of the X-direction component■-X-direction position of the X-direction component■+X-direction position of the y-direction component■-X-direction position of the y-direction component■+X-direction position of the X-direction component■-X direction of the X-direction component Position■+X-direction position of y-direction component■-X-direction position of y-direction componentIn the case of ■~■, it is the value of each direction position when the origin is set to the lower left, or in this case, the value near the origin is multiplied In order to prevent it from becoming 0, the values Q9 to Q16 of the position in each direction when trying to change the origin and set the origin to the right are also described in the same way.
It is assumed that a total of 16 values from Q1 to Q+e represent the length, direction, and position of each stroke segment of the target character.

In the conventional calculation method, it can be expressed as, for example, equation (1). In other words, this calculation method uses the character width (H in the case of formula (1))
Since the normalization is performed using X), it is possible to effectively absorb differences in the shape of each partial pattern due to individual differences.

(Problems to be Solved by the Invention) However, the conventional character recognition method using partial pattern Q values based on character width normalization has the following problems.

A sub-pattern of approximately vertical or approximately horizontal straight lines, e.g.
Regarding de° and “-”, the partial pattern Q value changes greatly due to its subtle slope (partial pattern Q
The value is + in addition to the X and Y components.

- Because the components are differentiated). Figures 6(a) to (Q) show explanatory diagrams of the problems in character width normalization in the case of the almost vertical straight line partial pattern "1". If a)) is normalized by the character width, the partial pattern Q value will be equivalent to that shown in (b) of the same figure.Here, the registered pattern stored in advance is slightly tilted to the right as shown in (c) of the same figure. By normalizing this pattern by the character width, it becomes the same Q value as that shown in (d) in the same figure.In other words, the matching result of these two patterns ((b) and (d) in the same figure) will be quite different.

Furthermore, if the pre-stored registered pattern is the average of several patterns, this will also not match the input partial pattern because different patterns tilted to the left and right are averaged. Will.

In both of these examples, the human cover turn (a) is
The desired matching result cannot be obtained because it is determined to be close to the intervening pattern shown in FIG. 2(e). In this way, conventional calculation methods can absorb differences in shape due to individual differences by normalizing based on character width, but this does not give good results in the case of partial patterns such as those mentioned above. There is a problem.

The present invention solves the above-mentioned problems and provides a character recognition method that performs matching using partial pattern Q values obtained by appropriate normalization for almost vertical or almost horizontal straight line partial patterns. purpose.

(Means for Solving the Problems) In order to solve the above problems, the present invention includes a preprocessing unit that removes unnecessary data from a coordinate data string obtained by manually writing on a tablet and performs linearization processing. , an extraction section that extracts feature points representing the characteristics of strokes that make up a written character from the coordinate data string linearized by the preprocessing section, and extracts each stroke for each partial pattern from the feature points obtained by the extraction section. In a character recognition method that recognizes human characters by determining a partial pattern Q value representing the characteristics of a segment and matching the partial pattern Q value with a pre-registered partial pattern Q value, an almost vertical or almost horizontal straight line is used. For the partial pattern, create a square whose side is the larger of the partial pattern's width in the X direction and the width in the Y direction, and perform square normalization to place the partial pattern in the center of the frame. The partial pattern Q value is determined based on the coordinate values of the pattern.

(Function) In the character recognition method of the present invention, X
Square normalization is performed in which the partial pattern is placed in the center of a square frame whose side is the larger of the width in the direction and the width in the X direction. That is, this square normalization acts to normalize the slope of the partial pattern without changing it. Therefore,
It is possible to eliminate the problem that the characteristic of inclination is not reflected in character width normalization in which the coordinate values of a partial pattern are normalized by dividing them by the character width.

Therefore, appropriate matching can be performed using the partial pattern Q value obtained by performing square normalization on the partial pattern as described above.

(Example) First, the concept of square normalization used in the character recognition method of the present invention will be explained. FIGS. 4(a) to 4(d) are explanatory diagrams of square normalization. FIG. 5A shows an example of a partial pattern of a substantially vertical straight line, which is slightly inclined to the right.

The partial pattern obtained by conventional character width normalization is shown in FIG. 4(b), and the partial pattern obtained by square normalization according to the method of the present invention is shown in FIG. 10(c). In conventional character width normalization, normalization is performed by dividing the coordinate values of the human-powered part pattern by the character width in the X and Y directions of the human-powered part pattern, that is, HX and HY, as shown in Figure (d). The result is the same as in the case of extremely sloped subpatterns. That is, although character width normalization has the feature of absorbing differences in shape due to individual differences, it cannot deal with patterns where the slope itself is characteristic. On the other hand, in the square normalization, a square frame is created using the larger of the character widths in the X and Y directions of the human-powered partial pattern, and the partial pattern is placed in the center of the square frame (FIG. 4(C)). That is, square normalization has the feature of reflecting the characteristic of a pattern whose slope itself is characteristic. Therefore, it can be said that square normalization compensates for the weaknesses of character width normalization.

Next, a specific calculation method using square normalization will be explained in comparison with character width normalization. FIG. 5 shows an explanatory diagram thereof. In the figure, 51 is a square normalization frame, and 52 is a character width normalization frame. Also, point A is the origin of the character width normalization method, point B is the origin of the square normalization method, XMIN, x are the X-direction coordinate values of point A and point B in the absolute coordinate system, HX, HY
are the width in the X direction and the width in the Y direction, respectively (HY>HX).

In the figure, since HY>HX, square normalization is applied only to the X direction. The conversion of point P in the character width normalization method is X, , =x-XM I N . Also, the transformation in the square normalization method is
m+Δx=x−XMIN+ΔX. Here, ΔX is determined by the following formula.

Δx = (HY - HX) / 2 In other words, the coordinate value X' of point P in the character width normalization method. is calculated as X', = xlI, /HX.

Also, the coordinate value X5゛ of point P in the square normalization method is x5°=X, /HY (the character width in the X direction is HY,
). The same result is obtained in both cases regarding the Y-direction coordinate value.

Y3°=Y'. = (y-YM I N)/HY In summary, the coordinate values obtained by the square normalization method are as follows.

However, if HY>HX, ΔX, Δy.

5HABA is Δx= (HY-HX)/2 Δy=Q SHABA=HY becomes. On the other hand, if HX>HY, ΔX.

Δy, 5HABA is ΔX=0 Δy=(HX-HY)/2 SHABA=HY becomes.

Next, an embodiment of the character recognition method of the present invention using the partial pattern Q value obtained by square normalization described above will be described.

FIG. 1 is a block diagram showing an example of the configuration of a character recognition device to which the method of the present invention is applied. In the figure, 1 is a tablet, 2 is a preprocessing unit that removes unnecessary data from a coordinate data string obtained by manually writing on the tablet, and performs linearization processing, and 3 is a coordinate linearized by the preprocessing unit. This is a feature point extracting unit that extracts feature points representing characteristics of strokes that make up a written character from a data string. A normalizing section 4 normalizes the partial pattern obtained from the output data of the feature point extracting section 3, and is composed of a character width normalizing section 4a and a square normalizing section 4b. 5 is a recognition unit, and a partial pattern Q value calculation unit 5 calculates a partial pattern Q value based on the output data of the normalization unit 4.
a, and a matching section 5b that matches the calculated partial pattern Q value with the partial pattern Q value calculated and registered in advance.

FIG. 2 shows an example of the structure of a character dictionary used for matching. Here, a character dictionary is a character and the information that separates the character (
This is a dictionary consisting of a partial pattern (called a cut position) and each partial pattern (partial pattern code). The matching method calculates how closely the input characters are divided at the cut positions defined in the character dictionary.

Here, it is assumed that the "*" attached to the upper right of the partial pattern is a designation to calculate the Q value of the partial pattern by square normalization.

Next, the operation of the characteristic portion of this embodiment will be explained using the flowchart shown in FIG.

First, it is assumed that the word "chu" is written on the tablet 1 as a human-powered character. Let's consider the first definition, ``phantom'' (Figure 2). Here, the cut position of "phantom" is (3, 1), so if the human characters are divided at that position, it becomes ",""1".Next, the mouth part patterns "fire" and "fu" of "phantom" are Since both are not designated as square normalization, partial pattern Q value calculation by character width normalization is performed by character width normalization unit 4a and partial pattern Q value calculation unit 5a.
(St, S2b.

S3). Next, the matching section 5b performs matching between the calculated partial pattern Q value and the partial pattern Q value of the intervening pattern, and calculates the distance between the partial pattern Q values and sets the wrinkles as an "illusory" distance (S4 ).

The cut position of the next definition “pull” is also similar to “phantom” (3, 1
), so if it is divided, the result will be the same (Figure 2). Here, "g calculates the partial pattern Q value by character width normalization in the same way as above (St, S2b, S3). On the other hand, "
1'' is a square normalization designation, so the calculation of the partial pattern Q value by square normalization is performed by the square normalization unit 4b and the partial pattern Q value calculation unit 5a (Sl, S2a).

S3). A matching unit 5 performs matching between the partial pattern Q value of these results and the partial pattern Q value of the registered pattern.
Calculate the distance of the partial pattern Q value by row b, and use it as “
The distance is set to "1" (S4).

Here, even if the input character "1" for "chu" is slightly tilted to either side, it will have a value close to the registered pattern.

In this way, if you decide whether to process character width normalization or square normalization on the character dictionary, you can effectively calculate the partial pattern Q value even for straight, vertical and horizontal partial patterns, and the result Therefore, the distance between the partial pattern Q values becomes smaller.

In the above embodiment, square normalization was specified by marking partial patterns, but various other methods are possible. For example, there may be a method of determining whether or not to process based on a partial pattern code, a method of determining based on the shape of an input pattern, etc. That is, it goes without saying that the present invention is not limited to the above-mentioned embodiment in the method of specifying square normalization.

As described above, according to this example, good results were not obtained when calculating the partial pattern Q value by normalizing the character width in straight, vertical and horizontal patterns. By performing square normalization in which a frame is created and a pattern is placed in the center, its characteristics can be reflected, and as a result, it is possible to reduce the distance due to matching of partial pattern Q values.

(Effects of the Invention) As described in detail above, according to the present invention, by performing square normalization on a substantially vertical or substantially horizontal straight line partial pattern, it is possible to reflect the characteristics of the slope of the partial pattern. Therefore, appropriate matching can be performed using the partial pattern Q value obtained by this square normalization.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a character recognition device according to an embodiment to which the method of the present invention is applied, FIG. 2 is a block diagram of a character dictionary used for matching in this embodiment, and FIG. Figure 4 is a diagram explaining the concept of square normalization, Figure 5 is a diagram explaining coordinate value calculation by square normalization, and Figure 6 is a diagram explaining the problems of conventional character width normalization. It is an explanatory diagram. 1...Tablet, 2...Pretreatment section, 3.
... Feature point extraction section, 4... Normalization section, 4 a -
--Character width normalization processing section, 4 b--Square normalization processing section, 5... Recognition section, 5a
... Partial pattern Q value calculation section, 5b... Matching section.

Claims (1)

[Scope of Claims] A preprocessing unit that removes unnecessary data from a coordinate data string obtained by handwritten input on a tablet and performs linearization processing; an extraction unit that extracts feature points representing the characteristics of the constituent strokes; a partial pattern Q value representing the characteristics of each segment of each stroke is determined for each partial pattern from the feature points obtained by the extraction unit; In a character recognition method that recognizes an input character by matching a value with a pre-registered partial pattern Q value, for an almost vertical or almost horizontal straight line partial pattern, the width in the X direction and the width in the Y direction of the partial pattern are Creating a square frame with the larger width as one side, and performing square normalization in which the partial pattern is placed in the center of the frame, and determining the Q value of the partial pattern based on the coordinate values of the partial pattern obtained. A character recognition method featuring