JPH0465786A - Method for recognizing hand-written character in drawing reader - Google Patents

Abstract

PURPOSE:To improve the recognition rate of an inclined character string having character skew by finding out the existence of oblique deformation from the size relation of respective totals of character string thinning vectors before and after normalization, removing skew due to the oblique deformation and recognizing each normalized character. CONSTITUTION:A hand-written character string in a read drawing is extracted and the sum of thinned vector lengths of each character is calculated. The coordinate transformation of the thinned vector of each character is executed by a prescribed normalizing transformation method for removing oblique deformation skew, the sum of thinned vector lengths of each character which consist of coordinate transformed vectors is calculated, the total vector length before normalization is compared with that after normalization, and when the total vector length is shortened by the normalization, the existence of oblique deformation is decided and skew due to the oblique deformation is removed to normalize each character. Then, respective characters in the normalized character string are recognized. Consequently, an inclined character string having character skew can be surely recognized.

Description

[Detailed Description of the Invention] [Summary] The purpose of the present invention is to improve the recognition rate of diagonal character strings with character distortion in a handwritten character recognition method for a drawing reading device that recognizes handwritten character strings written in reading drawings. The present invention is configured to detect the presence or absence of diagonal deformation in a character string extracted from a read drawing, remove distortion due to diagonal deformation, normalize each character, and then recognize each character. Further, the presence or absence of diagonal deformation is determined from the magnitude relationship between the sums of character string thinning vectors before and after normalization.

[Industrial Application Field] The present invention relates to a handwritten character recognition method for a drawing reading device that recognizes handwritten character strings written in a read drawing, and in particular, to a method for recognizing handwritten characters written in a read drawing by generating vector data from image data of a read drawing. The present invention relates to a handwritten character recognition method for a drawing reading device.

In recent years, in fields such as plant design and electrical design, systems that automatically input reading data of handwritten drawings into CAD systems are being put into practical use. It is limited to character strings aligned in the degree direction, and the recognition rate is low for character strings tilted at angles other than 0 degrees and 90 degrees, so character recognition confirmation and correction processing or re-input in the CAD system is required. there were.

Therefore, it is necessary to reliably recognize slanted character strings.

[Prior Art] Conventionally, as a recognition method for character strings tilted at angles other than 0 degrees and 90 degrees, there is a known recognition method for character strings tilted in 15-degree increments ("Identification of tilt in drawings"). (See Proceedings of the 37th National Conference of the Information Processing Society of Japan, 2W-2, 1988, pp. 1608-1609).

[Problems to be Solved by the Invention] However, in such conventional character recognition methods that target character strings tilted in 15-degree increments, each character constituting the character string is This method is effective for characters with no diagonal deformation, but if the character string is not erected and has diagonal deformation, characters cannot be recognized correctly due to this diagonal deformation distortion.

For example, when creating a piping diagram for plant design, character strings are entered by multiple people on one drawing, but the characters that make up the character string may be distorted due to the inclination of the character string. The presence or absence of such distortion differs from person to person, and it is necessary to detect diagonal distortion of characters forming a character string by taking individual differences into account.No effective solution has been found to date.

The present invention has been made in view of these conventional problems, and provides a character string recognition method for a drawing reading device that can reliably recognize diagonally tilted character strings with character distortion and improve the recognition rate. The purpose is to provide.

[Means to solve the problem] FIG. 1 is a diagram explaining the principle of the present invention.

First, the present invention is directed to a handwritten character recognition method for a drawing reading device that recognizes a handwritten character string written in a read drawing. As shown in FIG. 1(a), such a handwritten character recognition method includes a first step 10 of extracting a handwritten character string from a scanned drawing; a second process 12 for detecting; a third process 14 for normalizing each character by removing the distortion caused by the diagonal deformation when a diagonal deformation is detected in the second process 12; and a fourth step 16 for recognizing each character of the character string that has been deformed or that has been determined to have no oblique deformation in the second step 12.

Here, the method for detecting diagonal deformation of a character string in the second process is as follows: A first process of calculating the thinning vector length of each character constituting the character string; a second step of calculating the sum; and a second step of coordinately transforming the thinning vector of each character using a predetermined normalization transformation formula for removing oblique deformation distortion, and calculating the length of the thinning vector of each character formed by the coordinate transformation vector. 3 steps; A fourth step of calculating the sum of the thinning vector lengths of the normalized characters calculated in the third step; and the total vector length before normalization calculated in the second step and the fourth step. Compare the total vector length after normalization obtained in , ■ If the total vector length becomes shorter due to normalization, it is determined that there is diagonal deformation, and ■ If the vector length becomes longer due to normalization, it is determined that there is no diagonal deformation. It is characterized by comprising a fifth step of determining; and;

The normalization conversion formula used in the third step of this oblique deformation detection method is that the inclination angle of the character string is α, the starting point of the character string is (xo
3'o), any vector coordinate value to be normalized as (
x, y), ■=x-cos (α)XhXK+ (α)y=y
−sin (α)xhxK+ (a).

Also, the coefficient used in the normalization conversion formula is the starting point of the string (
x0, y0) is the length of the perpendicular to the straight line with slope α, h=l −sin (a) X (x−x0) +C
Coefficient K(α) given by O8(α)×(yy0) and used in the normalization conversion formula
is a value in the range of 1 to 0 set based on experience.

Furthermore, the coefficient Kl (α) of the normalization conversion formula used in the method for detecting diagonal deformation of character strings is 0<|α|≦π/
4, K1(α)=1.0π/4〈1α1≦3π/
8, K1(α)=0.73π/8<|α|≦π/2
When , K1(α)=0.0.

Furthermore, after determining the diagonal deformation in the fifth step of the method for detecting diagonal deformation of a character string, the total vector length before normalization multiplied by the weighting coefficient by 2 (α) and the total vector length after normalization are determined. and ■When the total vector length is equal or shorter due to normalization, it is determined that there is diagonal deformation; ■When the total vector length is longer due to normalization, it is determined that there is no diagonal deformation; .

The value of 2(α) for the weighting coefficient in this case is determined from experience as follows: when 0<|α|<=π/4, k ((2)=1.0+CO1(α/2)π/4<1α
When 1〈=3π/8, k (a) = 0. 7+cos (π/8)3π/
When 8<|α|<=π/2, k(α)=0.0.

[Operation] According to the handwritten character recognition method of the drawing reading device according to the present invention having such a configuration, as shown in FIG. 1(b), when a character string with an oblique deformation is detected, A normalized character string is generated according to a normalization conversion formula that removes character distortion due to diagonal deformation, and recognition processing is performed on the normalized character string without deformation distortion. The column recognition rate can be greatly improved.

As a result, the number of eyepiece diagrams used in plant design, etc. is 3.
It is possible to accurately recognize character strings in an eyelid diagram, including character strings arranged at an inclination of 0 degrees and 60 degrees, and to obtain a recognition rate sufficient to put automatic input into a CAD system into practical use.

[Embodiment] FIG. 2 is a block diagram of an embodiment of a drawing reading device in which the handwritten character recognition method of the present invention is implemented.

In FIG. 2, reference numeral 18 denotes a drawing reading unit, which reads hand-drawn drawings of plant designs and the like using, for example, an image scanner and converts them into image data. 20
is a character string extracting unit, which cuts out a character string written with an inclination from the image data of the drawing read by the drawing reading unit 18, and extracts a character string written with an inclination α and a cutting area set as a parallelogram. , a contour vector for each extracted character, and a thinning vector obtained by thinning the contour vector of each character. Reference numeral 22 denotes a diagonal deformation detection unit, which detects the presence or absence of diagonal deformation by comparing the sum of the thinning vector lengths before normalization and the sum of the thinning vector lengths after normalization, which will be explained later. . Reference numeral 24 denotes a character normalization unit, which normalizes each character using a predetermined normalization conversion formula for a character string determined to have a skew deformation by the skew deformation detection unit 22, specifically converts each character. Perform normalization transformation of the constituent thinning vector coordinates. 26
is a character recognition unit that refers to a dictionary and performs character recognition for characters without oblique deformation.

FIG. 3 is a flowchart of character recognition processing in the embodiment of FIG. 2.

In Fig. 3, character strings are first extracted using Sl from the image data of the reading drawing, that is, character strings existing in the reading drawing are extracted, and the slope α of the character string, the parallelogram cutting area, and each character are Generate contour vectors and thinning vectors.

Next, in S2, it is determined whether the inclination α of the character string extracted in 81 is 0° or other than 90°.

S3 if the inclination α of the character string is other than 0° or 90°
Proceeding to step S4, a process for detecting whether or not there is a diagonal deformation is performed, and only for character strings for which a diagonal deformation has been determined, character normalization processing is performed to remove distortion due to the diagonal deformation,
Finally, at step 85, character recognition is performed for characters without oblique deformation.

Next, the oblique deformation detection process in S3 and the character normalization process in S4 in the character recognition process flowchart of FIG. 3 will be described in detail.

FIG. 4 is a diagram showing the principle of detecting diagonal deformation according to the present invention. Taking the character string rAAJ as an example, the character string 28 with diagonal deformation, the character string 30 normalized to the character string with diagonal deformation, and the normalized character string A character string 32 obtained by further normalizing the character string 30 is shown. That is, the character string 28 has a diagonal deformation, and when this character string 28 is normalized to remove distortion due to the diagonal deformation, a character string 30 without any diagonal deformation is obtained. On the other hand, if the character string extracted from the reading drawing has no diagonal deformation as shown in the character string 30, by normalizing the character string 30 without diagonal deformation, the character string that has been distorted due to the diagonal deformation can be Column 3
It will be converted to 2. Case 2 is a case where the character string 28 with a diagonal deformation is normalized to a character string 30 without a diagonal deformation, and case ■ is a case where the character string 30 without a diagonal deformation is normalized to a character string 32 with a diagonal deformation. ■Let it be.

For the character string 28 before normalization and the character string 30 after normalization in case ■, by calculating the sum of the lengths of the thinning vectors that make up each character and comparing the magnitude relationship, in the case of the character string 28 with diagonal deformation, is in a relationship that the total vector length after normalization becomes longer.

On the other hand, when the character string 30 without diagonal deformation in case ■ is normalized, there is a gap between the sum of the thinning vector lengths before normalization and the sum of the thinning vector lengths after normalization, as shown in the figure. There is a relationship in which the total vector length after normalization becomes longer. Characters to be processed are determined based on the magnitude relationship of the sum of the thinning vector lengths of the characters that make up the character strings before and after normalization for character strings with and without such diagonal deformation. It is possible to detect whether a string, that is, a character string before normalization has a diagonal deformation.

FIG. 5 is a flowchart of a specific diagonal deformation determination process based on the diagonal deformation detection principle shown in FIG.

In FIG. 5, ``First, the thinning vector length of each character constituting the character string is 11. Calculate. For example, if the character string "A...Z" with the slope α shown in FIG.
, consists of arms.

Character thinning vector length l2. is, 1+1= az Cz + bz dz
)2. Here, i is a number indicating the order of each character constituting the character string, and j is a number indicating the number of thinning vectors constituting one character.

In the case of the character rAJ in Figure 6, for example, the thinning vector length 11□ of the first thinning vector Vll is the starting point coordinates
ll+ b 11), the end point coordinates are (C++, d++
) is calculated using equation (1) above. The thinning vector length is calculated using the thinning vector V1. of the first character rAJ. Thinning vector of the last character rZJ from ■7
, up to.

Next, the process proceeds to S2, and the total vector length of the character string rA-ZJ is calculated by m m(i) L=ΣΣ 1 1·I j=1.

Next, the process proceeds to 83, where the thinning vector V of each character is coordinate-transformed using the following normalization conversion formula, and the thinning vector 1. seek.

As shown in Figure 6, the coordinate transformation using equation (3) is as follows: α is the inclination angle of the character string, and the coordinates of the starting point of the character string are (x0, y
0), coordinate transformation for distortion correction by diagonal deformation is performed on arbitrary coordinate values (x, y).

Here, the constant in equation (3) is the length of the perpendicular to (x, y) before normalization with respect to the straight line with slope α that passes through the starting point (xo, yn) of the character string, as shown in Figure 7. and h=l −sin (α) X (x−Xo)
+cos (a)x(y y0)I (
4) Given as. Also, Figure 7 shows (text) after normalization.

y) is also shown. Furthermore, the value of +(α) in the coefficient in the conversion equation (3) is a value determined depending on the inclination angle α of the character string, and is preferably set empirically, for example, as follows. That is, as shown in FIG. 8, when the inclination angle α of the character string is less than π/2, correction is performed as K1(α) 1.0, and in the range where the inclination angle α exceeds π/2 and is less than 3/8π. In this case, K2(α) is set to 0.7 to suppress the correction ratio due to the sine and cosine components of the angle α, and K1(α)−0,0 is set at 3/8π or more, so that no distortion correction is performed.

Referring again to FIG. 5, in S3, the coordinates of the vector vlI are transformed according to the equation (3), and the normalized vector V,
Once , the process proceeds to S4, where, as in the case of Sl, the thinning vector length 11. of each character consisting of the conversion vector vlI is calculated. The process then proceeds to S5, where the total vector length for the converted character string is calculated in the same manner as in the case of 82.

Next, proceed to 86, and before conversion (before normalization) calculated in S2
The magnitude relationship between the total vector length and the total vector length after conversion (after normalization) calculated in S5 is compared. Specifically, the total vector length after conversion and the total vector length before conversion are compared with a value obtained by multiplying the coefficient determined by the inclination angle α of the character string by 2(α). Then, according to the conditional expression, the total vector length after conversion is equal to 2(α)
If it is shorter than XL, proceed to S7 and determine that there is oblique deformation,
If it is longer, the process proceeds to S8 and it is determined that there is no oblique deformation.

The value of 2 (α) for the coefficient of equation (6) used in S6 may be set empirically. For example, the diagonal deformation is determined to be present by the diagonal deformation determination process shown in FIG. If so, the process advances to S4 in FIG. 3, where the character string is normalized. The normalization of this character string can be performed using the above-mentioned equation (3) in determining the oblique deformation. Specifically, at 83 in FIG. 5, a conversion has already been performed to remove the distortion of the thinning vector Therefore, use the conversion result. On the other hand, when it is determined that there is no diagonal deformation in the diagonal deformation determination process shown in FIG. 5, the character string is processed as shown in FIG. Proceeding directly to S5, character recognition is performed.

In the above embodiment, the coefficients K2(α) and 2
Regarding (α), according to the empirical side, for example, equation (5), (7
), but the present invention is not limited to these empirical values, and appropriate coefficient values may be used as necessary.

Next, the recognition rate of the handwritten character recognition method of the present invention will be explained in comparison with the conventional method.

Figure 9 shows character strings of input drawings with different inclinations used to verify the character recognition rate. Regarding the input character strings in Figure 9, recognition of character strings with an inclination of 15 degrees is targeted. The recognition results of the conventional method are shown in FIG. In FIG. 10, characters marked with a circle indicate successful recognition, and characters marked with an cross indicate a failure of recognition. 1st
In the case of figure 0, the correct answer is 3 out of 55 characters targeted for recognition.
There were 0 characters and 25 characters were incorrect, and the recognition rate was approximately 54.5%.
Met.

FIG. 11 shows the recognition results obtained by the handwritten character recognition method of the present invention for the input character string shown in FIG.

In the case of the present invention shown in FIG. 11, among the 55 characters targeted for recognition,
The correct answers were 47 characters and the errors were 8 characters, achieving a recognition rate of approximately 85.5%, confirming a significant improvement in recognition rate compared to conventional methods.

[Effects of the Invention] As described above, according to the present invention, it is possible to reliably perform character recognition of drawings including handwritten character strings tilted at 30° or 60°, such as plant claw diagrams, etc. By improving the character recognition rate, it is possible to construct a drawing input system connected to a drawing reading device that can automatically input drawing data into a CAD system at a practical level.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of the principle of the present invention; Fig. 2 is a configuration diagram of an embodiment of the present invention; Fig. 3 is a character recognition processing flow diagram of the present invention; Fig. 4 is a diagram of the principle of diagonal deformation detection of the present invention: Fig. 5 The figure is a flowchart of the oblique deformation determination process of the present invention; Figure 6 is an explanatory diagram of detecting oblique deformation of the present invention; Figure 7 is an explanatory diagram of parameters used in the normalization transformation of the present invention; Figure 8 is a diagram of the normalization of the present invention Figure 9 is an illustration of the input character string used to verify the recognition rate of the present invention; Figure 10 is an illustration of the recognition results of the conventional method; Figure 11 is the figure of the present invention. FIG. In the figure, 18: Drawing reading unit 20: Character string extraction unit 22: Oblique deformation detection unit 24: Character normalization unit 26 Two character recognition units

Claims (7)

[Claims] (1) In a handwritten character recognition method for a drawing reading device that recognizes a handwritten character string written in a read drawing, a first step of extracting a handwritten character string from the read drawing; a second step of detecting the presence or absence of diagonal deformation in the character string; a third step of normalizing each character by removing the distortion due to the diagonal deformation when the diagonal deformation is detected in the second step; A fourth step of recognizing each character of the character string normalized in the third step or the character string determined to have no oblique deformation in the second step; Character recognition method. (2) In the handwritten character recognition method for a drawing reading device according to claim 1, the detection of the diagonal deformation of the character string in the second step includes the step of calculating the thinning vector length of each character constituting the character string. A second step of calculating the sum of the lengths of the thinning vectors calculated in the first step; Coordinate transformation of the thinning vector of each character using a predetermined normalization transformation formula for removing oblique deformation distortion; a third step of calculating the thinning vector length of each character formed by the coordinate transformation vector; a fourth step of calculating the sum of the thinning vector lengths of the normalized characters calculated in the third step; The total vector length before normalization calculated in the process is compared with the total vector length after normalization obtained in the fourth process, and when the total vector length becomes shorter due to normalization, it is determined that there is oblique deformation. , a fifth step of determining that there is no oblique deformation when the vector length becomes longer due to normalization; and a handwritten character recognition method for a drawing reading device. (3) In the handwritten character recognition method for a drawing reading device according to claim 2, the normalization conversion formula used in the third step is such that the inclination angle of the character string is α, the starting point of the character string is (x_0, y_0 ), the arbitrary vector coordinate values to be normalized are (x, y), and the vector coordinate values after normalization are (■, ■), then ■=x-cos
(α)×k(x)×K_1(α)■=y-sin(α)
A handwritten character recognition method for a drawing reading device, characterized in that it is given by ×k(x)×K_1(α). (4) In the handwritten character recognition method for a drawing reading device according to claim 3, k(x) used in the normalization conversion formula is a straight line with an inclination α passing through the starting point (x_0, y_0) of the character string. It is the length of the perpendicular line, k(x) = |-sin(α)×(x-x_0)+cos
(α)×(y−y_0)| Constant K_1 used in the normalization conversion formula
A handwritten character recognition method for a drawing reading device, characterized in that (α) is set to a value in the range of 1 to 0 based on experience. (5) In the handwritten character recognition method for a drawing reading device according to claim 4, the constant K_1(α) is determined based on experience, and when 0<|α|≦π/4, K_1(α)=1.0π/ 4
When <|α|≦3π/8, K_1(α)=0.73π/
A handwritten character recognition method for a drawing reading device, characterized in that when 8<|α|≦π/2, K_1(α)=0.0. (6) In the handwritten character recognition method for a drawing reading device according to claim 2, the determination of the diagonal deformation in the fifth step is performed using a value obtained by multiplying the total vector length before normalization by a weighting constant K_2(α). Compare the total vector length after normalization, and if the total vector length is equal or shorter due to normalization, it is determined that there is oblique deformation,
A handwritten character recognition method for a drawing reading device, characterized in that when the total vector length becomes longer due to normalization, it is determined that there is no oblique deformation. (7) In the handwritten character recognition method for a drawing reading device according to claim 6, the value of the weighting constant K_2(α) is determined from experience to be 0<|
When α|<=π/4, k(α)=1.0+cos(α/2) When π/4<|a|<=3π/8, k(α)=0.7+cos(π/8 ) A handwritten character recognition method for a drawing reading device, characterized in that when 3π/8<|α|<=π/2, k(α)=0.0.