JPH01197890A - Inclined character recognizing system - Google Patents

Abstract

PURPOSE:To surely recognize segment characters even when the characters are inclined characters by a contour scanning method by scanning scanning lines lines for the segment characters in vertical and horizontal directions so that the scanning directions can be parallel with rectangular coordinates inclined by a detected inclination angle for reading rectangular coordinates. CONSTITUTION:In the case of the inclined characters, a character string extracting part 16 retrieves line data 26 adjacent to the inclined character string, detects the inclination theta of the characters from the line data 26, and stores the detected inclination theta in a character string data table 18. Therefore, an edge table preparing part 22 can know the inclination theta of the characters from the character string data table 18, and the edge table preparing part 22 prepares an edge table by scanning the scanning lines by the contour scanning method in the vertical and horizontal directions so that the scanning directions can be parallel with the rectangular coordinates inclined at the inclination thetafor the reading rectangular coordinates. Consequently, even when the characters are the inclined characters, the same edge table as the one for the characters having no inclination can be prepared. Thus, even the inclined characters can be surely recognized.

Description

[Detailed Description of the Invention] ta] Regarding a method for recognizing slanted characters by reading and recognizing slanted characters drawn on handwritten drawings, the present invention aims to reliably recognize even slanted characters using a contour scanning method. Later, the inclination angle of the character is detected from the line data adjacent to the vectorized character string data, and when creating an edge table using the contour scanning method, the scanning line for the cut out character is read and the orthogonal coordinate is tilted by the detected inclination angle with respect to the orthogonal coordinate. Create an edge table by scanning vertically and horizontally parallel to the coordinates.

[Industrial Field of Application] The present invention relates to a slanted character recognition method for reading and recognizing slanted characters drawn on handwritten drawings.

In a figure recognition system that reads hand-drawn drawings such as maps, plant diagrams, circuit diagrams, etc. and converts them into data, character recognition processing is performed to recognize characters drawn on the drawings at the same time as figure recognition. is converted into data and stored in a file.

That is, when recognizing handwritten drawings such as maps, plant diagrams, circuit diagrams, etc., processing such as identification of eight line types, filling, area extraction, and information reduction (vectorization of binarized data) is necessary. At the same time, character recognition is also an important process, and in the future, character recognition will continue to be given great importance in the field of graphic recognition processing.

[Prior Art] In a character recognition method used in such a graphic recognition system, it has been considered to carry out the recognition process shown in FIG. 13, for example.

First, as shown in step S1, a drawing is read from an image scanner and converted into binary image data, and this binary image data is converted into a-line image data with a width of 1 pixel in step S2, and then in step S3 The thinning data is converted to vector data to compress the amount of information.

Next, in step S4, character string extraction processing is performed to extract data that is considered to be a character or character string based on the connectivity, structure, etc. of the vector data.

From the data obtained in this character string extraction process, step S5
Cut out the binary image data character by character in step S6.
Create an edge table using the contour scanning method.

The edge table created by this contour scanning method is a data table showing the continuity of pixels forming a character.

For example, the creation of an edge table using the contour scanning method for 2t[image data shown in FIG. 14] will be specifically explained as follows.

The edge tables created by the contour scanning method are the upper and lower edge tables that show the continuity of characters in the vertical direction (vertical direction) shown in Figure 15, and the upper and lower edge tables that show the continuity of characters in the horizontal direction (vertical direction) shown in Figure 1'6.
There are two edge tables, left and right, that indicate continuity in the horizontal direction.

The upper and lower edge table in FIG. 15 shows the starting point between the scanning line and the character (
It is created using y-coordinate data indicating the lower edge) and the end point (upper edge).

Here, the X marks in FIG. 14 indicate the upper and lower edges for creating the upper and lower edge tables.

For example, at the first scanning position (X-1) in FIG. 14, the lower edge (start point) is y=1, and the upper edge (end point) is y
=4, edge data "1" and r4J are stored in the upper and lower edge table of FIG. That is, the characters start at V=1 and end before y=4.

Furthermore, for example, for scanning position x=10, "1 to 3"
Edge data divided into three parts, ``6 to 8'' and 15 to 17'', is stored.

Similarly, the left and right edge table in Figure 16 shows the left edge (starting point) and right edge (end point) of the characters in Figure 14 obtained by scanning scanning lines in the y-axis direction at a predetermined pitch parallel to the X coordinate. The edge data (x coordinate value) of is stored. Note that the 0 marks in FIG. 14 indicate the left and right edges for creating the left and right edge tables.

In this way, if an edge table is created for all extracted characters in step S6 of FIG. 13, character features are extracted from the edge table in step S7, characters are recognized by comparing them with a dictionary, and the recognition results are Output.

[Problems to be Solved by the Invention] However, 1. The problem with the process of recognizing characters from an edge table using such a contour scanning method is the recognition of so-called slanted characters written at an slant in handwritten drawings. It is.

In other words, in the current contour scanning method, as shown in Fig. 14, the scanning direction of characters is vertical and horizontal directions parallel to the reading orthogonal coordinates, so the characters are tilted with respect to the reading orthogonal coordinates. Even if an edge table was created by scanning slanted characters, the characteristics of the characters extracted from the edge table of slanted characters were completely different from those in the dictionary, and there was a problem that slanted characters could not be recognized. .

The present invention was made in view of these problems, and
To provide a recognition method for slanted characters that can surely recognize even slanted characters written on handwritten drawings by a contour scanning method.

[Means for solving problems] FIG. 1 is a diagram explaining the principle of the present invention.

In FIG. 1, a handwritten drawing is read by a reading unit 10, converted into binary image data by a thinning processing unit 12, for example, into thinning data with a width of 1 pixel, and further converted into vector data by a vector conversion unit 14. , character string extraction unit 16
given to.

The character string extraction unit 16 extracts character strings from the vectorized image data, and creates a character string data table 18 and a vector data table 2 for each character based on the extracted character strings.
Create 0.

When the character string data table 18 and the vector data table 20 are created, the edge table creation unit 22 scans the characters determined by the vector data of each extracted character using the contour scanning method to create an edge table. Next, the recognition unit 24 extracts character features from the edge table and compares them with a dictionary to recognize the characters.

When writing slanted characters on a drawing, a leader line or the like is generally drawn adjacent to the slanted character string along the direction in which the characters are arranged, as shown in FIG. 1(b).

Therefore, in the case of slanted characters, the character string extraction unit 16
searches the line data 26 adjacent to the character string, detects the inclination angle θ of the character from this line data 26, and stores it in the character string data table 18.

Therefore, the edge table creation unit 22 can know the inclination angle θ of the character from the character string data table 18.
An edge table is created by scanning the test line using the contour scanning method in the vertical and horizontal directions in parallel with the orthogonal coordinates, which are tilted at an angle of θ with respect to the orthogonal coordinates. Create.

[Operation] According to the slanted character recognition method of the present invention, characters written at an slant on a handwritten drawing are horizontally and vertically determined using a coordinate system in which the scanning line of the character is tilted by the slant angle using the contour scanning method. Because of the parallel movement in the direction, it is possible to create the same edge table for slanted characters as for non-slanted characters, and even slanted characters can be reliably recognized based on the edge table.

[Embodiment] FIG. 2 is a block diagram showing an embodiment of the present invention.

In FIG. 2, reference numeral 10a denotes an image scanner as a reading section, which uses a line sensor or the like in which a plurality of light receiving pixels are arranged in a direction perpendicular to the drawing feeding direction, and outputs a video signal corresponding to graphic characters. Image scanner 10a
The video signal from is converted into binary image data by the A/D converter 28, and is stored in the buffer memory 30 as 2-digit image data for one drawing.
Valued image data is stored. The binarized image data of the read drawing stored in the buffer memory 30 is processed by the thinning processing unit 1.
In step 2, line thinning processing is performed to convert the image data into image data having a width of, for example, one pixel.

Note that although the buffer 7 memory 30 stores binary image data of figures and character strings, the processing after the thinning processing section 12 will be described with reference to only the binary image data of character strings.

The binary image data of the character string that has been converted into a to1 line by the thinning processing unit 12 is converted into vector data by the solid line conversion unit 14. The conversion into vector data in the vector conversion unit 14 is a process of obtaining coordinate data (vector data) for each pixel of the binary image data of the thinned character string in the reading orthogonal coordinates determined by the image scanner 10a. be. By vectorizing this binarized image data, data compression can be performed to reduce the amount of data to be handled in subsequent recognition processing.

The character string data vectorized by the vector conversion unit 14 is given to the character string extraction unit 16, which extracts data that is considered to constitute a character string from the vector data, and extracts the extracted character string data. (vector data), a character string data table 18 is created by a character string data table creation unit 18a provided in the character string extraction unit 16, and a vector data table 20 for each character is created by a character vector data table creation unit 20a. be done.

Here, when creating a character string data table in the character string data table creating section 18a, the inclination angle θ of a character is detected from line data adjacent to the character string and stored in the character string data table 18.

When a character string data table 18 and a vector data table 20 for each character are created in the character string extraction unit 16, the edge table creation unit 2 uses these table data.
2 creates an edge table for each character included in the character string using the contour scanning method. When creating an edge table using this contour scanning method, in the case of slanted characters, orthogonal coordinates that are inclined by the inclination angle θ with respect to the reading orthogonal coordinates are set based on the inclination angle θ stored in the character string data table 18.
An edge table of a slanted character is created by scanning a scanning line in the vertical and horizontal directions parallel to the orthogonal coordinates set based on the slant angle θ.

When the edge table creation unit 22 creates an edge table for each character, the character recognition unit 24 extracts character features from the edge table, matches them with a dictionary prepared in advance, and outputs recognition based on the matching results. come to arise.

Next, the oblique character recognition process according to the embodiment of FIG. 2 will be described in detail with reference to the operational flow diagram of FIG. 3.

First, in step S1 in FIG. 3, a handwritten drawing is read and scanned by the image scanner 10a, and binary image data of a character string drawn on the handwritten drawing is obtained by the A/D converter 28 and stored in a buffer memory. Write to 30.

Here, the handwritten drawing to be read by the image scanner 10a is, for example, a plant drawing as shown in FIG. A character string 32 is drawn, and a lead line 56a is usually drawn close to this slanted character string 32. In the slanted character recognition process of the present invention, line data such as the leader line 26a drawn close to the slanted character 32 is used as information for detecting the slant angle θ of the character string. Therefore, regarding how to write handwritten drawings that are subject to the recognition processing of the present invention, it is sufficient to decide in advance that for slanted characters, a leader line must always be drawn close to the direction in which the slanted characters are arranged.

Referring again to FIG. 3, the character string read by the image scanner 10a in step S1 is converted into thinned image data of, for example, 1 pixel width by the thinning processing unit 12 in step S2, and further converted into vector data in step S3. Conversion part 1
4, the thinned image data is converted into vector data.

Subsequently, in step S4, a character string data table 18 including the inclination angle θ of the character string is created for the vector data of the character string extracted by the character string extraction unit 16.

For example, if character string data as shown in FIG. 5 is obtained, the line data 26 is detected from the vector data located near the character string 32, and the character The inclination angle θ of column 32 is detected and stored in the character string data table.

FIG. 6 is an explanatory diagram showing an example of the character string data table 18 created in step S4 of FIG. 3.

That is, in the character string data table 18, for example, taking the character string data shown in FIG. , the height of the character string (H), the length of the character string (shi), the inclination of the character string detected based on the line data 26 (θ), the number of characters in the character string (N), the characters in FIG. The upper left and lower right coordinates of each character included in column 32 (for example, for character A, the upper left coordinates (xl, yl) and lower right coordinates (x2, V2)), and the vectors that make up the character A pointer to the data table is stored.

When the character string data table 18 based on the character string extracted from the vector data is created in step S4 in this way, a vector data table 20 for each character included in the character string is created in the next step S5.

FIG. 7 is an explanatory diagram of the vector data table 20,
The coordinates of the end point on the From side close to the y-axis side of the line data constituting each character included in the character string 32 shown in FIG. If the coordinates are stored as vector data and the line data is represented by curve approximation or the like, the fact that it is curve approximation is stored as additional information.

The vector data table 20 shown in FIG. 7 will be specifically explained using the character rAJ in the character string 32 in FIG. 5 as an example, as shown in FIG. 8.

That is, since the character rAJ is a combination of three line data, the From side end point F for the left side of the character rAJ is
1 coordinates and the coordinates of the To side end point T1 are stored, and the horizontal -
Regarding the line data, Fr.

The coordinates of the m-side end point F2 and the coordinates of the To-side end point T2 are stored, and furthermore, lr ro is stored for the line data on the right side of the character rAJ.
The coordinates of the m-side end point F3 and the coordinates of the To-side end point T3 are stored.

Such a vector data table contains the remaining characters r2J
r3J and r4J are created in the same way. As for the characters "2" and "3", since they have curved parts, vector data obtained by dividing the curve into multiple parts and linear approximation is stored, or coordinate data of appropriate end points is stored, and this end point is It is sufficient to store the curve data connecting between the two as attached information.

Referring again to FIG. 3, once the vector data table 20 for each character is created in step s5, the process proceeds to step S6, where an edge table is created for the vector data table for each character by the contour scanning method. When creating this edge table, the inclination angle θ of the character string stored in the character string data table in step S4 is used to determine the scanning direction of the scanning line.

Figure 9 shows the first character rA included in the character string 32 in Figure 5.
FIG. 10 is an explanatory diagram showing the creation process of the top and bottom edge tables taking J as an example, and FIG. 10 shows the creation process of the left and right edge tables of the character rAJ.

First, in the contour scanning process for creating the upper and lower edge tables shown in FIG. 9, the reading orthogonal coordinates (
xy coordinates), set the orthogonal coordinates (XY coordinates) that are tilted by the inclination angle θ of the character string with respect to the read orthogonal coordinates with the lower left end point of the character rAJ as the origin, and then The scanning line 34 is moved in parallel in the X-axis direction at a predetermined pitch α, and the value of the Y coordinate indicating the intersection of the scanning line 34 and the character rAJ is stored as edge data in the upper and lower edge tables.

Figure 11 shows the character r obtained by the contour scanning method in Figure 9.
The upper and lower edge tables of AJ are shown.

Incidentally, the pitch α of the scanning lines 34 in FIG. 9 is determined by the resolution calculated from the pixel density at the time of reading by the image scanner.

On the other hand, in the contour scanning for creating the left and right edge table shown in FIG. 10, as in the case of FIG. (XY coordinates) is set as the lower left end point of the character rAJ as the origin, and the scanning line 34 parallel to the X axis is moved in parallel in the Y axis direction at a predetermined pitch α, and the scanning line 34 and the character rAJ
Create a left and right edge table that stores the values of the X coordinates indicating the intersection with - FIG. 12 shows the left and right edge table 1q obtained by the contour scan of FIG. 10.

When creating an edge table for recognizing slanted characters in this way, it is necessary to create an edge table that is orthogonal to the reading coordinates based on the slant angle θ of the character string obtained from line data adjacent to the character string. Top and bottom and left and right edge tables are created by horizontal and vertical scanning according to the coordinates, and it is possible to create an edge table that has exactly the same contents as when the characters are not tilted.

Referring again to FIG. 3, once the edge table for each character has been created in step S6, character features are extracted from the edge table in the next step S7, compared with a dictionary prepared in advance, and recognized. The result will be output.

On the other hand, in the conventional character recognition process shown in FIG. 13, an edge table was created from the extracted binary image data, whereas in the present invention, the edge table The feature is that an edge table is created from vector data, and the detection of continuity of line data constituting characters, which was a circuit when creating an edge table from binary data, is now possible from vector data. This can be done more reliably by edge table detection, which can improve the character recognition rate.

[Effects of the Invention] As explained above, according to the present invention, even if the characters are slanted drawn on a handwritten drawing, an edge table that is exactly the same as when the characters are not slanted can be created using the contour scanning method. Even oblique characters can be reliably recognized by the contour scanning method.

[Brief explanation of the drawing]

Figure 1 is an explanatory diagram of the principle of the present invention; Figure 2 is a configuration diagram of an embodiment of the present invention; Figure 3 is an operational flow diagram of the recognition process of the present invention; Figure 4 is a plant to be read by the present invention. Explanatory diagrams of the drawings; Figure 5 is an explanatory diagram of slanted character string extraction data; Figure 6 is an explanatory diagram of a character string data table; Figure 7 is an explanatory diagram of a character vector data table; Figure 8 is an explanatory diagram of the character rAJ An explanatory diagram of the vector data table of: Fig. 9 is an explanatory diagram of creating the upper and lower edge tables using the contour scanning method: Fig. 10 is an explanatory diagram of creating the left and right edge tables using the contour scanning method: Fig. 11 is obtained from Fig. 9. An explanatory diagram of the upper and lower edge tables: Fig. 12 is an explanatory diagram of the left and right edge tables obtained from Fig. 10: Fig. 13 is an operation flow diagram of conventional character recognition processing: Fig. 14
The figure is an explanatory diagram of edge table creation using the contour scanning method: Figure 15 is an explanatory diagram of the upper and lower edge tables obtained by the contour scanning of Figure 14: Figure 16 is the left and right edge table obtained by the contour scanning of Figure 14. FIG. In the figure, 10: Reading section 10a: Image scanner 12: Thinning processing section 14: Vector conversion section 16: Character string extraction section 18 Two character string data table 18a: Character string data table creation section 20: Vector data table 20a: Vector Data table creation unit 22: Edge table creation unit 24: Recognition unit 26: ta data 26a: Leader line 2B: A/D converter 30: Buffer memory 32: Slanted character string 34: Scanning line y From Qin! January IF1/? , 1 day on the moon 1st figure 4ζ departure date 1-1 Fumi way 19th 1st earth JζC21 2nd figure O on II month ^! n work 〇-work 3 figure brushito drawing dI RfL mouth Monthly figure 4 father Ding 9・'hand I! 1st generation data theory January episode Figure 5 Explanatory diagram of vertical table outside the character string Figure 6 Figure 8 Dark brown 12 figure Figure 13: 10,000 k is a month's rotation Figure 14

Claims (3)

[Claims] (1) A reading unit (10) that reads handwritten drawings such as plant diagrams and circuit diagrams; a thinning processing unit (12) that thins the binarized image data obtained from the reading unit (10); and a thin line. a vector converter (14) that converts the binarized data into vector data; a vector converter (14) that extracts character strings from the vector data and creates a character string data table (16) and a vector data table for each character (
18); and a character string extraction unit (20) that reads the character data extracted character by character from the vector data table (18), scans the character data horizontally and vertically with scanning lines parallel to the orthogonal coordinate axes, and extracts the character data from the vector data table (18). an edge table creation unit (22) that creates an edge table from vector data of character intersection points obtained by scanning; and a recognition unit (24) that recognizes characters based on registered data of the edge table; The column extractor (16) detects the inclination angle (θ) of a character from the line data adjacent to the character string and stores it in the character string data table (18), and the edge table creator (22) The inclination angle (θ) with respect to the orthogonal coordinates
A method for recognizing a tilted character, characterized in that the edge table is created by scanning the scanning line in the horizontal and vertical directions of the character in parallel with the tilted orthogonal coordinate axis. (2) The character string data table (18) contains the upper left coordinates of the character string (A, B), the lower right coordinates of the character string (C, D), the height of the character string (H), and the length of the character string ( L), the inclination angle (θ) of the character string, the number of characters in the character string (N), and the upper left and lower right coordinates of the character cutting area are stored as character string data. Recognition method for slanted characters described in section. (3) The slanted character according to claim 1, wherein the vector data table (20) stores the starting point coordinates and ending point coordinates of line data constituting the character as vector data for each character. recognition method.