JP3446794B2 - Polygon approximation method and character recognition device for character recognition - Google Patents

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to character recognition,
In particular, the handwriting of numbers and hiragana or the like at the time of character recognition character
The present invention relates to a polygon approximation method for correcting an image and a character recognition device.

[0002]

2. Description of the Related Art Image of handwritten characters such as numbers and hiragana
In particular, in order to recognize handwritten characters and automatically perform clear writing and editing of these handwritten characters, many character recognition methods using a computer device have been conventionally proposed.

Examples of character recognition methods include those that use the contribution to the external direction and relaxation matching methods. In these methods, binary images in character units in which position and size are normalized after photoelectric conversion are used. For the data, a feature vector based on the degree of contribution to the outer contour or a line segment feature based on the relaxation matching method is created and matched with the dictionary table to classify the character images . Here, the outer-direction contribution is white (background) to black (character) when a character image is scanned from multiple directions.
It is composed of outline features that represent the overall features of the character by finding the points that change to (outline points), and direction contributions that indicate the local direction of the character due to the connecting lengths of the black (character) part in multiple directions. The feature selected from the directional contribution features at the outline points represented by the outline feature is the feature of the character image. Regarding the outline direction contribution, the Institute of Electronics and Communication Engineers 1983/10 Vol. J66-D
No. 118 of 10 (Norihiro Ogita, Seiichiro Naito, Isao Masuda)
See pages 5 to 1192 entitled "Identification of Handwritten Kanji Characters by Contribution Characteristics of Outer Direction".

On the other hand, in the relaxation matching method, after extracting the contour of a normalized character image in units of one character, polygonal approximation is performed using a large number of line segments to coordinate the start and end points of each line segment. This is a method for classifying character images by obtaining line segment features consisting of length and angle and matching with a dictionary table.

In the conventional character recognition method based on the relaxation matching method described above, when the recognition target is expanded to a total of 3036 character types of JIS 1st level Kanji 2071 type and Hiragana 65 type, a high recognition rate of about 95% is obtained. can be obtained, similar characters due to expansion of the character types (only part of a character such as "strange" or "make" is different, part of the after is the same character type) due to an increase in, the character image It is said that the dictionary line segment and the line segment feature of the read character do not correspond well to the crushed or missing characters that occur (especially frequently when handwritten characters are read), which may result in erroneous recognition. There was a problem.

A method disclosed in Japanese Patent Application Laid-Open No. 63-175984 solves the problems of the conventional relaxation matching method. In this method, the boundary of the binary image is traced to find some closed point sequences, and cos θ is found for the angle θ formed by any three points in the section of the closed point sequence. However, it is assumed that the arbitrary three points are equal from the middle point to the other two points. Then, as a cutting method 1, if cos θ equal to or larger than a threshold value having a maximum value locally exists in the above section, the curve is cut at the midpoint and polygonal approximation of the boundary line is performed.
Next, as a cutting method 2, in order to obtain a characteristic of a bent portion having a large bending angle of a character, a cutting point obtained by the cutting method 1 is set as a starting point,
If the distance between the boundary curve of the input character that is the end point and its approximate curve is found, and if that distance is greater than or equal to the specified value, the point on the boundary curve that gives that distance is taken as the cutting point, and this is repeated to curve all boundaries The character image approximated by a polygon is obtained by setting the distance between the curve and the approximated curve within a specified value. FIG. 17 shows an example of a character image subjected to polygonal approximation by the polygonal approximation method (polygonal approximation image of a binary image of handwritten character “NO”).

[0007]

However, in the above method, the contour lines 1705 and 1706 of one of the writing lines and the contour lines 1710 and 1711 at the target position have different inclinations, so that a visually unnatural character image is formed. , In addition,
Since the thickness of the entry line is ignored, even though the entry line has a certain thickness, there is no line thickness (see FIG. 1).
7) (the portion within the dotted lines 1701 and 1701 of 7) occurs.

[0008] The present invention has been made to solve the above problems in the character recognition process, without ignoring the thickness of the entry line, the natural character images visually
An object is to provide a polygonal approximation method and a character recognition device to be obtained .

[0009]

In order to solve the above-mentioned problems, the polygonal approximation method of the present invention first traces the contours of binary image data and first defines a plurality of contour lines having information on the contour pursuit direction. If the feature of the specific first contour vector is the contour vector and the feature of the specific first contour vector is matched by the feature conversion means configured by associating the specific feature with the output feature, the specific first feature The second contour vector is obtained by converting the features of the contour vector into the output features, and the first contour vector that is not converted into the second contour vector is also the second contour vector. A feature is that the vectors are combined into a third contour vector, and the feature data of the second contour vector and the third contour vector that are not combined as the third contour vector are extracted. .

In the embodiment, the second contour vector that has not been combined as the third contour vector and the third contour vector having a length less than the specified length are determined as indeterminate vectors, and the second contour vector is determined as an indefinite vector. The indeterminate vector is deleted based on the information obtained from the feature data of the contour vector and the third contour vector, and the third vector before and after the indeterminate vector is deleted.
It is desirable to correct the contour vector or create the fourth contour vector, and extract the remaining feature data of the second contour vector, the third contour vector, and the fourth vector.

The character recognition apparatus of the present invention has a normalized 2
A contour pursuit means for tracing the contour of the value image data, a first contour vector determination means for extracting a plurality of contour lines having information on the contour pursuit direction as a first contour vector, and a specific feature and an output feature are associated with each other. When the feature of the specific first contour vector of the continuous first contour vector matches the specific feature by the template configured as described above, the feature of the specific first contour vector is converted into the output feature, and the second contour vector is converted. And a second contour vector conversion means for converting the first contour vector not converted into the second contour vector into the second contour vector, and the second contour vector continuous in the same direction in the order of contour tracing to synthesize a third contour vector. A second contour vector synthesizing means for obtaining a contour vector, and a second contour vector and a third contour vector which are not synthesized as a third contour vector.
The present invention is characterized by comprising polygonal approximation processing means for extracting the characteristic data of the contour vector.

Furthermore, in the embodiment, the polygonal approximation processing means further determines a second contour vector which have not been synthesized as a third contour vectors, the third contour vector below a defined length as undefined vector Based on the information obtained from the indeterminate vector determination means and the feature data of the second contour vector and the third contour vector, which are determined to be the indeterminate vector, the indeterminate vector is deleted, and the third contours before and after the indeterminate vector are deleted. It is desirable to have an indefinite vector correction unit that corrects the vector or creates the fourth contour vector, and extract the remaining feature data of the second contour vector, the third contour vector, and the fourth vector.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION 1. Character Recognition Device FIG. 1A is a block diagram showing the configuration of an embodiment of a character recognition device 10 using the polygonal approximation method of the present invention.
Is a photoelectric conversion unit, 12 is an image memory, 13 is a normalization preprocessing unit, 14 is a normalization processing unit, 15 is an identification unit, 16 is a polygon approximation processing unit, 17 is a recognition unit, and 18 is a dictionary.
It should be noted that in FIG. 3, a hardware configuration such as a control circuit (CPU) and an internal memory as a computer device and a control line are not shown. Further, when the polygonal approximation method of the present invention is applied to a character recognition device or the like, the polygonal approximation processing means 16 may be constructed by hardware, but it may be constructed by a program, ROM, magnetic disk, floppy disk, etc. It is desirable that the control circuit (CPU) such as a character recognition device using a computer stores the data in the medium and executes it as part of the recognition process.

The photoelectric conversion unit 11 converts the input character pattern into an electric signal and outputs it to the image memory 12. The normalization pre-processing unit 13 binarizes each pixel data stored in the image memory 12 at an appropriate level as a pre-stage process of the character recognition process, or is represented by the pixel data.
The circumscribing frame of the character image (binary image data) is determined. The normalization processing unit 14 normalizes the position and size of the character image that has been preprocessed by the normalization preprocessing unit 13 such as movement, enlargement, and reduction, and cuts out the character image in character units. . The recognition unit 15 includes a polygon approximation processing unit 16 and a recognition unit 17. The polygonal approximation processing means 16 performs the polygonal approximation processing of the present invention (described later) on the character image of each character cut out by the normalization processing unit 14, and the characteristic data of the character image after the polygonal approximation processing. (Coordinates, direction, length of line segment start point) and line segment number data are given to the recognition means 17. The recognizing means 17 associates the character image after the polygonal approximation processing with the feature data registered in the feature data dictionary 18 and measures the distance based on the line segment number data, and the digitized distance is the best. The character types of the reference models that are close to each other are output as the recognition result of the character image . When the dictionary 18 is initialized, the photoelectric conversion unit 1 is used for each reference model of each character type.
The characteristic data and the line segment number of the reference model of each character type obtained through the conversion by 1, the preprocessing by the normalization preprocessing unit 13, the normalization by the normalization processing unit 14, and the character segmentation are substituted.

FIG. 1B is a block diagram showing an example of the configuration of the polygon approximation processing means 16. The polygon approximation processing means 16 is shown in FIG.
Is the first contour vector determination means 161 and the contour tracking means 16
1 ', a second contour vector converting means 162, and a third contour vector synthesizing means 163. It is desirable to add an indefinite vector determining means 164 and an indefinite vector correcting means 165 for correcting an indefinite vector (described later) and extracting the characteristic data of the line segment.

2. Polygon Approximation Processing FIG. 2 is a flowchart showing the basic operation of the polygon approximation processing means 16 during polygon approximation processing.

In step S1, polygon approximation processing means 16
Is the character image obtained by the preprocessing stage of image recognition.
Di, i.e., receives the normalized binary image data after the processing of each stage leading to the normalization process from the reading of handwritten characters described above, from the normalization processing unit 14.

Next, in step S2, the polygon approximation processing means 16 traces the contour of the character image (see the example of FIG. 7) received from the normalization processing portion 14 by the contour tracing means 161 ', and a plurality of first pluralities are obtained. (See the example in FIG. 8). The first contour vector determination means 161 is cut out by 2
The bitmap including the value image data is scanned from a predetermined contour tracking start position, for example, one line in the right direction from the upper left dot, and when scanning up to the rightmost, one dot down (that is, the next line) from the leftmost. Similarly to the right 1
By scanning for a line and repeating this method, the tracking direction is determined based on the contour tracking means 161 ′ by using the pixel above the first one of the black pixels that appears first as the pixel of interest, and the line segment vector (first contour vector ) Is created, and the data for each line segment vector, that is, the coordinates of the start point, the direction, the length, or the coordinates of the end point are stored in the memory. Further, based on the contour tracing means 161 ′, tracing is continued around the same direction, and a contour for one character consisting of a plurality of line segment vectors (first contour vector) is created. As a result, the data of the first contour vector, that is, the coordinates of the starting point, the direction, the length, or the ending point of each line segment vector are stored in the memory. At this time, contour tracing is performed in a fixed clockwise or counterclockwise direction determined by the contour tracing means 161 'used first. The operations of the first contour vector determination means 161 and the contour tracking means 161 'will be described later.
Here, when the character image to the binary image data first contour vector determining unit 161 has received are representations of images, such as "5" as shown in FIG. 7 contour is one, character type There may be a plurality of contours depending on the handwritten character reference pattern. For example, if the character image is something like "9" or "stone," the outline shows the outside and the closed inside shows two parts.
In the case of a "rice field", the outline shows the outside 1
There are five in total, four indicating the inner part and the closed inner part. In such a case, the first contour vector is created in order from the one that constitutes the outermost part of the image, and the creation of the first contour vector is completed. After that, it is checked whether there is a portion in the bitmap where the first contour vector has not been created, and if there is, a first contour vector is created for that portion.
For example, when the character image is "T", the contour data is composed of a set of five first contour vector data groups.

In step S3, the second contour vector conversion means 162 arranges the first contour vectors in the order of the contour tracing described above in order to correct the partial distortion of the character image .
A second contour vector is obtained by performing conversion based on a preset correction template (see the example of FIG. 10) (for example, FIG. 10).
Figure 10 shows a distorted lower left image like (a).
(The image is such that distortion such as'a 'is removed). Of the first contour vectors, the one that does not match any preset template is directly used as the second contour vector. The data of the created second contour vector (the coordinates of the starting point, the direction, the length, or the coordinates of the ending point) is stored in the memory. 10 (a) to (d) and (a) '
The correction template shown in the example of (d) ′ is a 3 × 3 template, but not limited to this, the resolution (pixel density) when the document is read and converted into an electric signal, the accuracy of the contour to be acquired, and the binary value. The template of m × n (m and n are natural numbers) is used according to the conditions in the previous stage of the polygonal approximation process such as normalization of image data.

Next, in step S4, the third contour vector synthesizing means 163 reads out each second contour vector from the memory and examines it in the order of contour tracing, and continuously examines the second contour vectors in the same direction.
Whether or not there is a contour vector is determined, and when there are two or more second contour vectors continuous in the same direction, these second contour vectors are combined (added) to form a third contour vector,
Control goes to step S5. If the two examined second contour vectors are not continuous in the same direction, the first one second contour vector of the two preceding and following second contour vectors is regarded as the third contour vector, and the process proceeds to step S5. .

In step S5, the data of the third contour vector and the regarded third contour vector (the coordinates of the starting point, the direction, the length, or the coordinates of the end point) are sequentially stored in the memory as feature data. It is desirable to associate line segment information (usually line segment number data) with the feature data or store it in a memory as part of the feature data. As described above, the operations of steps S4 and S5 are repeated until the second contour vector for one character is completed, whereby the polygonal approximation of the character image, that is, the received binary image data is performed.

3. Indeterminate vector correction processing In step S4 of FIG. 2, the continuous second contour vector in the same direction is examined, the continuous second contour vector in the same direction is combined as the third contour vector, and the second non-contiguous second contour vector is synthesized. The contour vector is regarded as the third contour vector, and the coordinates of the starting point, the direction, the length, or the coordinates of the end point are used as the characteristic data. However, in the correction processing of the indefinite vector, instead of this, the regarded third contour vector component and the Of the three contour vector components, the one whose length is less than the specified value is regarded as an indefinite vector, and is regarded as an indefinite vector.
Correction processing is performed on the contour vector and the third contour vector.

FIG. 3 is a flow chart of polygonal approximation processing which is the basic operation of the polygonal approximation processing means 16 of FIG. In FIG. 3, the operations of steps S1 to S4 are the same as those in FIG.

In step S5 ', the indefinite vector determination means 164 (FIG. 1 (b)) checks the length of the deemed third contour vector and the third contour vector, and determines the length of the deemed third contour vector and the third contour vector. If the value is less than the specified value is determined to be an indefinite vector, the process proceeds to step S6 if there is an indefinite vector, and the process proceeds to step S7 if there is no indefinite vector.

In step S6, conditions such as the position, direction, and length of the indefinite vector are checked, and correction processing corresponding to those conditions is performed (see the embodiment). One of the conditions for performing the correction process is that the total length of the indefinite vector is equal to or less than the specified value α, and the end point of the third contour vector immediately before the indefinite vector and the start point of the third contour vector immediately after the indefinite vector. Is less than or equal to the specified value λ, and in this case, the indefinite vector is deleted, the position and length of the third contour vector before and after the indefinite vector are adjusted, and The contour vector is converted into a fourth contour vector. As the second condition for performing the correction process, the total length of the indefinite vector is between two specified values β and γ, and
Regarding the angle θ formed by the direction of the indefinite vector and the direction of the fourth contour vector created by the correction, cos θ is a prescribed value δ
In the above case, it is, and in this case, the indefinite vector is deleted, and the fourth contour vector connecting the end point of the third contour vector immediately before the indefinite vector and the start point of the third contour vector immediately after the indefinite vector is newly added. To create. If the uncertain vector does not satisfy the conditions for performing the above two correction processes, the uncertain vector is set as the uncorrectable uncertain vector, and the process proceeds to step S7.

In step S7, the data of the third contour vector and the regarded third contour vector (the coordinates of the starting point, the direction, the length, or the coordinates of the ending point) are sequentially stored in the memory as the characteristic data when the process shifts from step S5 '. To do.
It is desirable that the characteristic data be associated with line segment information or be stored in the memory as a part of the characteristic data. Also,
When the process shifts from step S5 ′, the data of the third contour vector other than the indefinite vector and the regarded third contour vector, the fourth contour vector, the third contour vector corresponding to the uncorrectable indeterminate vector, and the regarded third contour vector The data of the three contour vectors are sequentially stored in the memory as feature data. It is desirable that the characteristic data be associated with line segment information or be stored in the memory as a part of the characteristic data.

4. First contour vector determination means, contour tracking means The first contour vector determination means 161 includes a plurality of 3 × 3 first contour vector determination templates 4 as illustrated in FIG.
The first contour vector is determined using 1 to 48, and the contour tracking means 161 ′ performs the contour tracking by obtaining the target pixel required for determining the first contour vector. The first contour vector determination templates 41 to 48 are adjacent to the target pixel C (meaning the central pixel of the matrix 3 × 3) and the target pixel C 8
It is composed of pixels. In FIG. 4, “0” indicates a white pixel and “1”
Represents a black pixel, and * means that the pixel at that position may be a white pixel or a black pixel. In this case, the templates 41 to 48 of FIG. 4 determine the direction of the first contour vector to be created, and which of the templates 41 to 48 is to be applied is determined by the position of the pixel of interest C. The direction of the first contour vector is determined by the direction of the template. In the case of FIG. 4, the following meanings (direction determination rules) are given to the templates 41 to 48 when the tracking direction is counterclockwise.

<Counterclockwise Direction Determination Rule> Template 41: The first contour vector to be created is set to the left. Template 42: The first contour vector to be created is directed downward. Template 43: The first contour vector to be created is oriented rightward. Template 44: The first contour vector to be created is directed upward. Template 45: The first contour vector to be created is directed to the lower left. Template 46: The first contour vector to be created is directed to the lower right. Template 47: The first contour vector to be created is directed to the upper right. Template 48: The first contour vector to be created is directed to the upper left. Note that although the target pixel C is a white pixel in FIG. 4, it may be a black pixel. However, when the target pixel is a white pixel, a contour vector circumscribing the character image is obtained, but each tracking direction determination template in FIG. 4 in which the target pixel is a black pixel is a contour vector inscribed in the character image. Used when seeking. Further, when the tracking direction is set to the clockwise direction, the clockwise direction determination rule in which the direction of the first contour vector is the reverse direction may be set for each template by the counterclockwise direction determination rule described above. Further, instead of the templates 41 to 48, another template may be created and given a meaning (direction of the first contour vector).
Further, as will be described later, the number of tracking direction determination templates is defined by the number of directions. Therefore, four directions are four tracking direction determination templates, eight directions are eight tracking direction determination templates, and sixteen directions are tracking direction determination templates. 16 templates are required.

Next, the contour tracing operation of the contour tracing means 161 'will be described with reference to FIGS. 5 and 6 and the template 41.
The description will be made using .about.48. The cut-out character image is an image 51 in FIG. As described in the description of step S2 in FIGS. 2 and 3, the first contour vector determining unit 161 moves to the right from the predetermined contour tracking start position 50 of the bitmap including the binary image data of the cut out character image 51. Scans one line in the direction, and scans to the rightmost one dot 1 dot below (that is, the next line)
Similarly, one line is scanned from the leftmost to the right, and this method is repeated, and the pixel that appears one dot above the first black pixel is set as the target pixel C (see FIGS. 5A and 5B). In addition,
5 (b) is a partially enlarged view of FIG. 5 (a)). Therefore, the tracking direction determination template applied first is the template 41 in which the pixel of interest C is located on the black pixel, and when the above-described counterclockwise direction determination rule is applied, the first approximate line segment (first The direction of the contour vector) is determined leftward.

The second and subsequent pixels of interest C are determined as follows. In FIG. 5, the presence or absence of dots in eight directions around the current pixel of interest C1 is checked from the direction of the first contour vector V1 determined by this pixel of interest C1. When the first contour vector is in the leftward direction, the search order matrix shown in FIG. 5C is assumed, and the position n satisfying the following two conditions is sequentially searched from the position of search order 0.

<Search Condition> The dot at the position of n (0 ≦ n ≦ 7) is a white pixel, and the dot of n + 1 is a black pixel. Therefore, in the case of FIG. 5A, when n = 1,
Since the condition of is satisfied, the pixel at this position (that is, the pixel 52 at the lower left of the current pixel of interest C1) becomes the next pixel of interest. Therefore, when the arrangement of the eight pixels adjacent to the pixel 52 is examined, there are black pixels to the right and below the pixel 52 as the pixel of interest, so the template 45 corresponds. Therefore, the direction of the second first contour vector V2 is determined as a vector pointing downward left as shown in FIG.

Next, when contour tracing is performed as shown in FIGS. 6A and 6B (first contour vectors V1 to V4 have been determined), the contour tracing means 161 'sets the pixel of interest at present. The pixel of interest next to C4 (FIG. 6B) is determined. Therefore, the presence or absence of dots in eight directions around the current pixel of interest C4 is checked from the direction (downward) of the first contour vector V4 determined by this pixel of interest C4. If the first contour vector is downward, the search order matrix shown in FIG. 6C is assumed, and the position n satisfying the following two conditions is searched in order from the position of search order 0. FIG. 6 (a),
In the case of (b), when n = 1, that is, the pixel 62 at the lower right of the pixel of interest C4 is the next pixel of interest that satisfies the above conditions.

Here, when the target pixel is determined, the direction of the first contour vector is determined, so that the direction of contour tracing is determined. Therefore, the method for determining the next pixel of interest is to set the pixel next to the current pixel of interest as position 0 as described above, and check the pixels of positions 1 to 7 clockwise if it is clockwise and counterclockwise if it is counterclockwise. Since the position satisfying the above conditions may be the next pixel of interest, the pixels of interest are sequentially obtained in this way in order to obtain the first contour vector of the entire character image 51 .
It is sufficient to stop the tracking operation after making one turn to the left.

[0034]

FIG. 7 shows an example of normalized binary image data . In this embodiment, a polygon image is approximated by applying the polygon approximation processing means 16 shown in FIG. 3 to the binary image data.
The process of obtaining ( FIG. 12) is shown in FIG.
This is shown in FIG.

FIG. 8 shows the second contour of FIG. 7 approximated by the first contour vector obtained by the first contour vector determining means 161.
It is a character image represented by the value image data, and the binary image data of FIG. 7 is obtained by the first contour vector determination templates 41 to 48 of FIG.

FIG. 9 is a diagram showing the correspondence between the direction of the contour vector and the numerical value indicating the direction used in each correction template of FIG. 10, and the numerical values 1 to 8 are associated with each direction. . Although the direction is set to 8 in the present embodiment, the direction is not limited to this, and the direction may be any multiple of 4, for example,
If there are four directions, assign a numerical value of 1 to 4 to each direction, and 1
If there are six directions, numerical values of 1 to 16 may be associated with each direction. Further, as described above, the number of first contour vector determination templates is defined by the number of directions (that is, the number of directions = the number of first contour vector determination templates).

FIG. 10 shows the second contour vector conversion means 162.
This is a correction template used in, and is used to correct a partial distortion of a character image . The second contour vector conversion means 162 arranges the first contour vectors in the order of contour tracking and converts them based on a preset correction template to obtain a second contour vector. For example, as shown in the correction template 101 of FIG.
Correction template partial images FIG 10 (a) '10
It is assumed that the partial image is one in which distortion such as 1'is removed. Template 101 shown in (a) to (d) of FIG.
To 104 are examples of matching templates to be matched with the first contour vector obtained in step S2 of FIG. 3 (FIG. 2),
Templates 101 'to 10' shown in (a) 'to (d)'
4'is a part that matches the matching templates 101 to 104
This is an example of a conversion template for correcting the minute image , the template 101 corresponds to 101 ', the template 102 corresponds to 102', and the same applies to the following templates.

FIG. 11 is a character image according to the polygonal approximation method of the present invention before the correction processing of the indefinite vector, and the second contour vector converted using the correction template of FIG. 5 and the second contour vector without using the correction template. It is a character image by the polygonal approximation method composed of the first contour vector which is the contour vector and the third contour vector synthesized (added) by the third contour vector synthesizing means 163, and the processing step of the polygonal approximation processing means 16 From the perspective of Fig. 3 (Fig. 2)
This corresponds to the character image obtained in step S4.
Note that, in FIG. 11, the contour vectors 111, 112, etc. are the first contour vectors that are the second contour vectors without using the correction template, and the contours 113 to 116, etc. are the third contour vectors.
The contour 117 is a contour vector, and the contour 117 is converted into the second contour vector by using the correction template as is apparent from the comparison with FIG.
It is the third contour vector synthesized (added) by 3.

FIG. 12 is a polygon approximation image after the correction processing of the indefinite vector of FIG. 3, and the contour vectors 111 and 112 existing in FIG. 11 by the correction processing of the indefinite vector.
It is easy to see as shown in FIG.
A character image was formed.

FIGS. 13 to 15 are views showing an embodiment of the correction processing of the indefinite vector in step S6 of FIG. FIG.
Is an example of performing the correction process according to the condition 1 for performing the above-described correction process. In FIG. 13A, the third contour vector 11 is placed immediately before the plurality of indefinite vectors 1102 surrounded by broken lines.
01, immediately after the third contour vector 1103, the total length of the two third contour vectors 1101 and 1103 is smaller than the specified value α, and the end point of the third contour vector 1101 and the start point of the third contour vector 1103 are Is the specified value γ
Smaller than In this case, the uncertain vector correction unit 165 deletes the uncertain vector 1102, and the third contour vector 11
The length is adjusted by changing the position of the end point of 01 and converted into the fourth contour vector 1201 shown in FIG. The indefinite vector correction means 165 further adjusts the length by changing the position of the end point of the third contour vector 1103, and converts it into the fourth contour vector 1202 shown in FIG. 13B.

FIG. 14 shows the condition 2 for performing the above-mentioned correction process.
This is an example of performing the correction process by
There is a third contour vector 1301 immediately before the plurality of indefinite vectors 1302 surrounded by broken lines, and a third contour vector 1303 immediately after that, and the total length of the indefinite vectors is the specified value β.
And γ, and cos θ is equal to or greater than the specified value δ with respect to the angle θ formed by the directions of all the indefinite vectors and the direction of the fourth contour vector created by the correction. In this case, the indefinite vector correction unit 165 deletes the indefinite vector 1302, and the third contour vector 130 is deleted as shown in FIG.
A fourth contour vector 1401 connecting the end point of 1 and the start point of the third vector 1302 is newly created.

FIG. 15 shows an example of the correction process when the normalized binary image data has a lot of noise and the first contour vector is disturbed.
502 and third contour vectors 1501 and 1 before and after 502
503 is shown. In the case of this example, since the conditions for performing the above two correction processes are not satisfied, the indefinite vector 1502 is not corrected and remains as it is as shown in FIG. 15B.

Next, the normalized binary image data of FIG.
Against shows an example of representation of a characteristic of the polygonal approximation character polygonal approximation contour vector of the image in FIG. 12 as an embodiment of a polygonal approximation character image according to the present invention (characteristic data). When (X coordinate of starting point, Y coordinate of starting point, direction, length) is used as the expression method of the contour vector, the polygon-approximated contour vector as shown in FIG. 1, 2, 6) (8, 7, 3, 10) (9, 17, 5, 6) (15, 18, 4, 2) (17, 20, 3, 4) (16, 24, 2, 3) ) (13,27,1,5) (8,26,8,2) (6,24,1,3) (3,25,3,3) (4,28,4,3) (7,31) , 5, 8) (15, 30, 6, 6) (21, 24, 7, 6) (20, 18, 8, 5) (15, 13, 1, 2) (13, 12, 7, 2) (14,10,5,14) (28,9,7,3) (27,6,1,13) (15,5,6,3) (18,2,7,2) ( 7, 0, 1, 3) and is expressed. However, it is assumed that the origin O is at the upper left and the direction follows the reference numerals in FIG.

FIG. 16 (a) shows the binary image data of the hiragana "no", and FIG. 16 (b) performs the processing of steps S2 to S6 of FIG. 3 on the binary image data of FIG. 18 (a). It is the figure which approximated the polygon. According to the present invention, as is apparent from FIG. 16, it is possible to realize a polygonal approximation that is natural in appearance and in which the thickness of the entry line is known.

[0045]

As is apparent from the above description, according to the present invention, a partial image of a first contour vector obtained by subjecting a normalized binary image to polygonal approximation processing is used .
Distortion or the like is corrected to obtain a second contour vector, a second contour vector that is continuous in the same direction is combined as a third contour vector, and polygonal approximation is performed. The contour vector and the third contour vector are set as indefinite vectors, and the ones that satisfy a specific condition among the indefinite vectors are corrected to further perform polygonal approximation. Therefore, the binary image data after the polygonal approximation process looks natural. A character image that realizes a polygonal approximation that allows you to see the thickness of the entered line with
Expressed as Ji .

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a character recognition device using the polygonal approximation method of the present invention.

FIG. 2 is a flowchart showing a basic operation of polygonal approximation processing means during polygonal approximation processing.

3 is a flowchart of polygonal approximation processing in which correction of an indefinite vector is added to the basic operation of the polygonal approximation processing means 16 of FIG. 2 during polygonal approximation processing.

FIG. 4 is an example of a tracking direction determination template.

FIG. 5 is an explanatory diagram of contour tracking.

FIG. 6 is an explanatory diagram of contour tracking.

FIG. 7 is an example of normalized binary image data.

FIG. 8 is the character image of FIG. 7 approximated by a first contour vector.

FIG. 9 is a diagram showing correspondence between a direction of a contour vector and a numerical value indicating a direction in a correction template.

FIG. 10 is an example of a correction template.

FIG. 11 is a polygonal approximation image of the present invention before correction processing of an indefinite vector.

FIG. 12 is a character image by the polygonal approximation method of the present invention after correction processing of an indefinite vector.

FIG. 13 is an example of correction processing of an indefinite vector.

FIG. 14 is an example of correction processing of an indefinite vector.

FIG. 15 is an example of correction processing of an indefinite vector.

FIG. 16 is a diagram showing an example of polygon approximation according to the polygon approximation method of the present invention.

FIG. 17: Polygonal approximation is performed by a conventional polygonal approximation method.
It is a figure which shows the example of an image .

[Explanation of symbols]

10 Recognition device 16 Polygonal approximation processing means 41 to 48 First contour vector determination template 111, 112 Second contour vector 113 to 117 Third contour vector 161 First contour vector determination means 162 Second contour vector conversion means 163 Third contour Vector synthesizing means 164 Indeterminate vector determining means 165 Indeterminate vector correcting means 1101, 1103, 1301, 1303 Third contour vectors 1102, 1302, 1502 Indeterminate vectors 1201, 1202, 1401 Fourth contour vectors C, C1 Focused pixels V1, V2, V3 , V4 first contour vector

Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06K 9/00-9/82 G06T 7/60 G06T 9/20

Claims (4)

(57) [Claims] 1. A feature conversion means configured to trace a contour of binary image data and use a plurality of contour lines having information of a contour pursuit direction as a first contour vector, and to associate specific features with output features. When the feature of the specific first contour vector for the first contour vector that matches the specified feature is converted to the output feature by converting the feature of the specific first contour vector to the output feature, The first contour vector that is not converted into the contour vector is also the second contour vector, the second contour vectors that are continuous in the same direction in the order of contour tracking are combined to form the third contour vector, and the first contour vector that is not combined as the third contour vector 2 contour vector and 3rd
A polygonal approximation method in character recognition, which is characterized by extracting feature data of a contour vector. 2. A further, a second contour vector which have not been synthesized as a third contour vectors, the third contour vector less than the length of the prescribed determination as undefined vector was determined to be indefinite vector, second contour vector Based on the information obtained from the feature data of the third contour vector and the third contour vector, the indeterminate vector is deleted, the third contour vector before and after the indeterminate vector is corrected, or the fourth contour vector is created. outline vector is extracted third contour vector, and the characteristic data of the fourth vector, it is characterized in claim
A polygonal approximation method in character recognition according to item 1. 3. A character recognition device, comprising a normalized 2
A contour pursuit means for tracing the contour of the value image data, a first contour vector determination means for extracting a plurality of contour lines having information on the contour pursuit direction as a first contour vector, and a specific feature and an output feature are associated with each other. When the feature of the specific first contour vector of the continuous first contour vector matches the specific feature by the template configured as described above, the feature of the specific first contour vector is converted into the output feature, and the second contour vector is converted. And a second contour vector conversion means for converting the first contour vector not converted into the second contour vector into the second contour vector, and the second contour vector continuous in the same direction in the order of contour tracing to synthesize a third contour vector. A second contour vector synthesizing means for obtaining a contour vector, and a second contour vector and a third contour vector which are not synthesized as a third contour vector.
A character recognition device comprising polygonal approximation processing means for extracting feature data of a contour vector. 4. The character recognition device according to claim 3,
The polygonal approximation processing means further includes a second contour vector that has not been synthesized as a third contour vector, an indefinite vector determination means that determines a third contour vector having a length less than a prescribed length as an indefinite vector, and the indefinite vector. Judged,
An indeterminate vector is deleted based on information obtained from the feature data of the second contour vector and the third contour vector, and the third contour vector before and after the indeterminate vector is corrected or a fourth contour vector is created. A character recognition device having a vector correction means and extracting the remaining feature data of the second contour vector, the third contour vector, and the fourth vector.