JP2917427B2 - Drawing reader - Google Patents

Description

Description: BACKGROUND OF THE INVENTION A. Field of the Invention The present invention relates to a drawing reading apparatus for detecting character data from a document or drawing in which characters, line segments, symbols, etc. are mixed, and recognizing the characters.

B. Summary of the Invention The present invention extracts character data from binarized pixel data corresponding to black and white obtained by raster-scanning a subject such as a document or drawing in which characters, line segments, and symbols are mixed. In a drawing reading apparatus for performing recognition, an outline vectorization processing unit for performing outline vectorization processing on the binarized pixel data, a registration unit for registering circumscribing rectangle data of character candidates in a tree structure, A character string candidate extraction unit that extracts a character string candidate from
A character string extraction unit that extracts a character string from the character string candidate; and an evaluation value that indicates a degree of approximation between the character candidate in the extracted character string and a character in a dictionary of alphanumeric characters and symbols. And a character cutout unit having a function of determining the type of a character string by comparing the maximum value (threshold value) of the evaluation value with a predetermined value, so that characters, line segments, and symbols are mixed. When reading and recognizing characters from documents and drawings, prevent erroneous recognition of kanji and alphanumeric characters and symbols, prevent character cutout errors, and use characters of any size and format. It is intended to be able to read easily and extremely accurately, and to simplify the processing.

C. Prior Art When reading characters and figures, for example, subjects such as documents and drawings are raster-scanned by an input device such as an image scanner and converted into black and white binary image data (input pattern). The character data of the object is extracted from the binary image data, and the object is recognized. In order to recognize a character included in a drawing or the like in which characters, graphics, and the like are mixed, it is necessary to first separate the character portion from the graphic portion, cut out the information for each character, and perform recognition.

D. Problems to be Solved by the Invention Here, the characters included in the drawings are not fixed in size and the format is not fixed, unlike the text. For this reason, it has been difficult to extract and recognize characters included in documents and drawings in which characters, line segments, symbols, and the like are mixed irrespective of the format and size. In particular, it has been extremely difficult to recognize handwritten sentences and the like having large fluctuations in character width and many separated characters. There is also a drawback that the processing takes time.

Further, in the conventional method of extracting data of one character from character string data based on assumed character width and character spacing, it is often useful for a character type having a constant character pitch and character width, In the case of handwritten characters or the like, erroneous clipping may be performed because the basis of the assumed character width and character spacing is thin and the basis is small. In the characters in the drawings, most of the characters are alphanumeric strings, but a small number of kanji character strings may be mixed. The following errors occur in a character cutout unit and a character recognition unit that attempt to correctly recognize such a drawing without distinguishing between alphanumeric characters and Chinese characters. That is, when characters are cut out from a character string as shown in FIG. 32 (a), numerals 1, 2, 3 and 4 are mistakenly integrated. Further, as shown in FIG. 32 (b), the English character "R" is erroneously recognized as the Chinese character "Shaku".

The present invention has been made in view of the above points, and an object of the present invention is to prevent characters from being erroneously recognized and erroneously cut out, and to easily and extremely accurately detect characters included in documents and drawings in any size and format. It is an object of the present invention to provide a drawing reading apparatus which can read and has a high processing speed.

E. Means for Solving the Problems The present invention extracts character data from binarized pixel data corresponding to black and white obtained by raster-scanning a subject in which characters, line segments, and symbols are mixed. A drawing reading apparatus that recognizes characters by referring to and comparing the character data and the character dictionary, based on binarized pixel data corresponding to black and white obtained by raster-scanning the subject. An outline vectorization processing unit that obtains outline vector data formed by combining vectors connecting black pixels and obtains as a character candidate based on circumscribed rectangle data circumscribing the outline vector, and an outline vectorization processing unit A registration unit for registering the circumscribed rectangle data of the selected character candidates in a tree structure, and determining a search range from a core of the character string candidates set to a predetermined size, and One character string candidate extraction unit that extracts as a character string candidate, from among the data extracted by the character string candidate extraction unit that center coordinates of the registered character candidates are present within the search range,
Character string extraction for extracting data in which the height h _c of the character string candidate nucleus and the distance d between adjacent character candidates in the character string candidate are d <h _c × k (k is a constant) as a character string When the projection in the horizontal direction of a plurality of adjacent character candidates in the character string extracted by the character string extraction unit overlap, a height direction integration function to integrate the plurality of character candidates, Each character candidate integrated by the function is compared with a dictionary of alphanumeric characters and symbols to perform character recognition, and in accordance with the plurality of recognition results, the degree of approximation between the character candidate and a character in the dictionary is determined. An alphanumeric character recognition function for obtaining each evaluation value to be represented;
For each of the alphanumeric character and symbol dictionary characters and symbols, a threshold value is set to a maximum value in an evaluation value sequence representing the degree of approximation between a plurality of samples of the same character and symbol and the dictionary characters and symbols. A character string type determination function that determines whether a character string is an alphanumeric character string by comparing the set threshold value with an evaluation value obtained by the alphanumeric character recognition function And the first deviation between the width of the character candidate determined not to be an alphanumeric character string and the assumed character width having a width equal to the size of h _c × k is close to the width of the character candidate. When the width of the character candidate and the size obtained by combining the distance between the character candidate and the nearby character candidate are larger than a second deviation between the assumed character width and the second deviation is smaller than a predetermined threshold value. Having a width direction integration function for integrating the character candidate and character candidates in the vicinity thereof, And a character segmentation unit which cuts out a character from the character string, is characterized by performing character recognition on the basis of the character data cut out by the character cut-out section.

F. Function When the circumscribed rectangle is obtained by the vectorization processing unit, a subject (document or drawing) in which characters, line segments, and symbols are mixed.
, Character candidate data is obtained. The circumscribed rectangle data of the character candidate is registered in the registration unit in a tree structure.
The character string candidate extraction unit searches data in the registration unit in a search range determined by the character string candidate nucleus, and extracts a character string whose center coordinates are within the search range as character string candidates. Since the circumscribed quadrangle data is registered in a tree structure, the neighborhood search processing of the area quadrangle is reduced. In the character string candidates,
Data satisfying the relationship of d <h _c × k (d is the distance between adjacent character candidates, h _c is the height of the character string candidate nucleus, and k is a constant) is extracted as a character string by the character string extraction unit. That is, line segments and symbols mixed in the subject are excluded, and only character string data is extracted. Next, the character cutout unit cuts out characters from the character string. First, when the horizontal projections of a plurality of character candidates that are close to each other in the height direction overlap in the character string, the character candidates are integrated. After integration in the height direction, each character candidate is compared with a dictionary of alphanumeric characters and symbols, and character recognition is performed. Then, an evaluation value is obtained corresponding to a plurality of recognition results. The evaluation value thus obtained is compared with a preset threshold value to determine whether or not the character string is an alphanumeric character string. If the result of the determination is that the character string is an alphanumeric character string, the width direction integration is not performed. If the character string is not an alphanumeric character, the following width integration is performed. That is, for example, the width of the character candidate is slightly smaller than the assumed character width, the width of the character candidate, the width of the candidate adjacent to the candidate, the size obtained by adding the distance between the character candidate and the adjacent character candidate, and the assumed character width. Is large. In this case, since the first deviation is smaller than the second deviation, the character candidates are not integrated. Further, for example, the width of the extracted character candidate is extremely narrower than the assumed character width, and the size of the width of the character candidate, the width of the candidate adjacent to the candidate, the size obtained by combining the distance between the character candidate and the adjacent character candidate, and the assumed character It is assumed that the difference from the width is relatively small. In this case, the first deviation is larger than the second deviation. At this time, if the second deviation is larger than a predetermined threshold, the character candidates are not integrated, but are integrated only when they are smaller than the threshold value. Then, the character candidates integrated in the horizontal direction and the height direction are cut out as one character. This allows characters to be read as characters regardless of their size or format.

In addition, the distinction between alphanumeric characters and kanji is made surely, and is not erroneously recognized. Furthermore, alphanumeric characters are not accidentally merged across the width.

G. Embodiment Hereinafter, an embodiment of the present invention will be described with reference to the drawings. First, a drawing reading apparatus according to the present invention includes an outline vectorization processing unit 1, a registration unit 2, a character string candidate extraction unit 3, a character string extraction unit 4, a character cutout unit 5, and a character recognition unit 6 as shown in FIG. Thus, the entire processing flow is shown in FIG. That is, first outline vector processing unit 1 performs the pre-processing in step S _1, the registration unit 2 in step S _2, the character string candidate extracting unit 3, a character string extracting unit 4 and the character segmentation unit 5 performs character cutout process, character recognition unit 6 performs character recognition processing in step S _3. The pre-processing performed by the contour vectorization processing unit 1 is represented by a flowchart as shown in FIG. That performs image input processing in step S _1, whereby the border vectorized process in step S _2, performs the component separation process in step S _3. Of these, the image input processing and the contour vectorization processing are performed as follows, for example, by the contour detection device described in the patent specification of Japanese Patent Application No. 63-78684.

FIG. 4 is a diagram showing a configuration of the contour detecting device, and 11 is a contour extracting unit. As shown in FIG. 5, the contour extraction unit 11 fetches binarized pixel data corresponding to black and white obtained by raster-scanning a subject such as a document or a drawing, and outputs pixel data of 2 × 2 pixels. Are sequentially extracted along the scan line, a command sequence for contour detection is created based on the pixel data, and output to the contour analysis unit 12 at the subsequent stage.
The contour analyzer 12 updates the records of the contour element table, the contour management table, and the contour connection table in the table storage 13 based on the command sequence. In addition, the contour analysis unit 12 uses the contour element table and the contour management table to extract, from the points constituting one contour pixel sequence, a point that is maximum when viewed from the start point (a macroscopic processing unit 31), and performs this processing. Are extracted from the relationship with the set threshold value at the point having the maximum distance between the two consecutive points extracted in step (microscopic processing unit 32), and the contour vector of the point extracted by this processing is extracted. Is described in the contour vector series table 33, and a short (short) vector of the contour vectors is removed by the short vector removing unit 34, and the description is updated.

Here, the contour element corresponds to a vector connecting two black pixels adjacent to each other, and a contour is formed by combining these. In the contour element table, as shown in FIG. 6, a unique code is assigned to each of the contour elements and the contour composed of these contour element groups, and for each contour element, the coordinates and directions and the contour elements belong. This is for describing the sign of the contour to be connected and the signs of other contour elements connected before and after the contour element, respectively. The contour management table is for describing the codes of the contour elements located at the front end and the end for each contour as shown in FIG. 7, and the contour connection table is for raster scan as shown in FIG. This is for describing the coordinates arranged in the direction, the sign of the unconnected contour element having the front end or the rear end at the coordinate, and the distinction between the front and rear ends of the unconnected end of the contour element. The contour vector sequence table is for describing the code of the start point coordinates for each contour vector sequence as shown in FIG.

The command string output from the contour extraction unit 11 includes information on the connection relationship between the pixel related to the pixel data of 2 pixels × 2 pixels extracted at that time and the contour pixel, and information on the black and white array pattern of the pixel data. And a coordinate code indicating the X coordinate of the pixel data. An example of the connection relationship is shown in FIG.
When pixel data of 2 pixels × 2 pixels surrounded by a bold frame in FIG. 9B is fetched, it becomes information that the front end of the contour element located at the X coordinate _Xn of the pixel data is connected. In this example, the coordinates of the pixel data, taking the coordinate pixel P ₁ located on the right under the paper surface. When the pixel data surrounded by the bold frame in FIG. 2B is fetched, the coordinates immediately before the X coordinate of the pixel data are obtained.
It becomes information that the rear end of the contour element located at X _n-1 is connected. In FIG. 10, a frame with a circle indicates a black pixel, and a frame without a circle indicates a white pixel. Focusing on, for example, the bold frame in FIG. 10 (a), the connection pattern code in this case includes the above-described connection information and black-and-white array pattern information in the bold frame. and the combination of the coordinate code indicating the X-coordinate of the pixel P _1. This coordinate code can also be generated on the side of the contour analysis unit 12 in synchronization with the timing of taking in the command sequence.

When the command sequence obtained in this way is taken into the contour analysis unit 12, the following processing is performed. Now, assuming that a command sequence relating to the pixel data of the thick frame in FIG. 11 is fetched, the contour element Cj indicated by the contact point is registered in the contour element table, and its direction and coordinates are entered.
The numbers of the other contour elements connected before and after Cj are entered in the connection element number column. In this case, since the contour element Cj is located in front of the contour element Cj, Ci is entered in the rear connection column of the contour element Cj column, and Ci is entered in the front connection column of the contour element Cj column. Further, in this example, the number of the contour to which the contour element Cj belongs is entered as Si in the contour number column. For the direction of the contour elements, for example 8-connected, vertically as shown in FIG. 12, left and right, a ₁ ~a ₈ corresponds to the diagonal direction is defined,
In the case of four connections, four directions, up, down, left and right, are defined.
Then, for the contour number Si of the contour management table, the leading contour element number column is updated from Ci to Cj, and for the X coordinate of the pixel data in the contour connection table, the forward connection column of the unconnected contour element number is Ci. From to Cj. By the way, in the actual processing, since the contour element to be connected to the contour element Ci is clarified by the contour connection table, the description of Cj in the front connection column related to the contour element number Cj column of the contour element table is the contour. This is performed by referring to the connection table. In the above description, when two contours generated separately with the progress of the raster scan are connected to form one contour, one of the belonging contour numbers is integrated into the other and becomes the same.

In the processing as described above, a point which is the maximum when viewed from the start point is extracted from the points constituting one contour pixel row, and a point having the maximum distance between the two extracted connected points is extracted. . The extracted contour vectors are stored in a table.

As described above, the input image of the subject, for example, the character “A” is represented by the outline vector as shown in FIG. 13 by the image input processing and the outline vectorization processing.

Next, an element separation process (step S3 in FIG. ₃ ) for extracting a character candidate will be described. First, short vectors are removed using the contour vector series table shown in FIG. FIG. 14 is a flowchart for explaining the short vector removal means, (shown in FIG. 15) from the outline vector sequence table shown in FIG. 9 at step S ₁ the circumscribed quadrangle of the contour vector (x _min, y _min), ( X _max , Y _max ) are calculated. In step _{S 2 dx = X max -X min} , dy = y max -y
Compare _min with threshold values DX _th and DY _th . Step S ₃ in the comparative determination of the step S2, in step S ₃ (dx> D
Determining X _th) ∪ a (dy> DY _th), to create information that circumscribed rectangle as a contour vector in the step S _4, if YES. An example of this information is shown in FIG. 16 as a circumscribed rectangle information table. Remove the vector of the same series in Step S ₅ from the outline vector sequence table shown in FIG. 9, if NO to remove noise (noise) in the step S _3.

FIG. 15 is an explanatory view showing a circumscribed rectangle of the contour vector. In this figure, x _max , x _min , y _max , and y _min are as follows.

_{x max = max (..., x} i, ...) x min = min (..., x i, ...) y max = max (..., y i, ...) y min = min (..., y i, ...) dx = x _max −x _min dy = y _max −y _min (dx <DX _th ) ∩ (dy <DY _th ) where DX _th and DY _th are threshold values.

The circumscribed rectangle information (13th
Character candidates are extracted under the following conditions based on FIG.

Condition w _x <threshold and w _y <threshold (w _x is the length of the circumscribed rectangle in the x direction, w _y is the length of the circumscribed rectangle in the y direction) Condition The outline vector of the outer periphery and the complete circumscribed rectangle of the character candidate The included outer and inner contour vectors are linked to contour vectors that have become character candidates.

As described above, the pre-processing is performed as shown in FIG. 3 to FIG.
Next, the character cutout processing (step S2 in FIG. ₂ ) will be described.

First, this character cutout processing is performed on the character candidates obtained by the contour vector processing unit 1 as shown in FIG. 17 in the horizontal direction (step S ₁ ), the vertical direction (step S ₂ ), and the diagonal direction (step S _3). ). The processing in each direction has the same contents, and after performing the character string candidate extraction processing (step S ₁ ) as shown in FIG. 18, the character string extraction processing (step S ₂ ) is performed, and then the character is cut out. processing (step S ₃₎
I do.

The processing of each step is performed as follows. First, in the character string candidate extraction processing, as shown in FIG. 19, a character candidate tree structure processing (step S ₁ ) is performed, and then a character string candidate range search processing (step S ₂ ) is performed. Actually 16th
As shown in the figure, the character candidate is registered in the registration unit (registration unit 2 in FIG. 1) as a tree structure in which spatial division into two is repeated based on the center coordinates of the circumscribed rectangle of the character candidate. A character candidate having a center coordinate in a search range as shown in FIG. 20 (a) is used as a core of a character string candidate having a height substantially equal to the height of the character string to be searched among the character candidates. Search from the character candidate tree of the registration unit 2. This search is performed by the character string candidate extraction unit 3 in FIG. 1 in the following procedure.

The center point of another character candidate within the search range as shown in FIG. 20 (b) is searched rightward from the character string nucleus.

Of the center points found above, the one located farthest from the central nucleus is the start point of the next search range. If not found at the center point of the search, the search in the right direction ends.

The same applies to the left direction.

Those found by the above search are set as character string candidates.
By performing a search in which the core size of character string candidates is set in several stages from the largest, character strings with different sizes can be handled.

Next, the character string extraction unit 4 extracts a character string that meets the following conditions from the character string candidates as a character string. That is, as shown in FIG. 21, when the height h _c of the character string candidate nucleus and the distance d between the character candidates have a relationship of d (i, j) <h _c × constant (1), I and j in the form of a character string.

Next, the character cutout unit 5 cuts out characters from the extracted character string in units of character candidates in a procedure as shown in FIG. First performed in the height direction separator integrated as step S ₁ in the case of matching the conditions shown in the following second equation (2).

min ｛xsi, xsj｝ ≦ (xsi or xsj or xei or xej) ≦ max ｛xei, xej｝… (2) (However, xs: x-direction start coordinate of circumscribed rectangle xe: x-direction end coordinate of external rectangle) Expression (2) is a conditional expression indicating that the widths of the character candidates i and j in the character string overlap, and the actual integration processing procedure is performed according to the flowchart of FIG. For example, as shown in FIGS. 23 (a) and 23 (b), when the character candidate i, j is separated in the height direction and the condition of the above-mentioned expression (2) is satisfied, the character candidate i, j is It is integrated as shown in FIG. Then the 22nd
In step S ₂ in FIG, the character candidate that is the height direction integration process the alphanumeric characters, and symbols of the dictionary are compared, character recognition is performed. Then, as shown in FIG. 26, a table of recognition results and their evaluation values is created for character string data as shown in FIG. 25 (character string after height direction integration processing). In FIG. 26, the data numbers are the character candidates C _{1 in} FIG.
A number of -C _5, sequentially arranged from those good degree of matching between the dictionary for each data, their evaluation values for each recognition result has also been tabulated. The evaluation value is a numerical value indicating how close the recognition result is to a character in the dictionary. The smaller the evaluation value is, the more similar the character is to the character in the dictionary. In next step S _3, compares the threshold value table created as FIG. 27 in advance by experiments, and the alphanumeric character recognition (Step S ₂₎ data table (Figure 26) created by , The type of the character string is determined.

Here, when the character data sample of "A" is recognized by a dictionary including kanji and alphanumeric characters, and the correct recognition result (first place) is obtained, that is, the evaluation value when the recognition result of A is obtained is e. _A. However, when an arbitrary character A is recognized, the evaluation value is not always eA even if the result is correctly recognized as _A. Therefore, the same experiment is performed on many character data samples of A.
If the evaluation series at that time is E _A , E _A = ｛e _A0 , e _A1 … e _An ｝
It is. If any character is recognized, the result is "A"
And if the evaluation value at that time is e _etc , then e _etc <ma
If x {E _A }, the result is valid. Therefore th _A threshold table shown in FIG. 27 is th _A = max
Define as {E _A }. The same kind of experiment was repeated for characters B to Z, 0 to 9, and symbols, and the maximum value E _th of the threshold table was changed to E
_th = max ｛th _{A to} th _Z , t _{h0 to} t _h9 , Defined as｝.

The determination of the character string type is performed according to a procedure as shown in FIG.
That compares the maximum value E _th of the first position of the string average e-mean value of the threshold table of FIG. 27 of the evaluation values of the recognition results in Figure 26 the table in step S _1. As a result, if the expression e-mean> E _th (3) is satisfied, since E _th is a value obtained by the above experiment, it is at least not an alphanumeric character string (that is, a kanji string). Can be determined. Also, e
If −mean ≦ E _th holds, it is determined whether or not the recognition result of each character is appropriate in steps S ₃ , S ₄ , and S ₅ on condition that the character to be determined remains (step S ₂ ). Is performed with reference to the threshold table shown in FIG. When the first of the recognition result of the character is a threshold table of Figure 27 with is a character having a character code of "A", to take the evaluation value th _A following numbers of recognition at that time Indicates that it must not. If the recognition result of a certain character does not satisfy the expression e <(corresponding threshold value table)... (4) for the evaluation value e corresponding to the result, the second or lower place of the recognition result It is checked whether or not there is one satisfying the condition of the expression (4). Such a determination is made for all character candidates of the character string, and if even one character does not satisfy Expression (4), it is determined to be a kanji character string.

Next, in step S ₄ in FIG. 22, the following first equation (5), when both matches the conditions shown in equation (6) is performed in the width direction separator integrated (Kanji column only).

(However, : Assumed character width wi: width of character candidate i wij: width of circumscribed rectangle containing character candidates i, j (ie,
Equation (5) indicates that the difference between the assumed character width and the width of the character candidate i is the difference between the assumed character width and the character candidate i, j. This indicates that the difference is larger than the difference between the included widths (i.e., the sum of the widths of i and j and the distance between i and j). Equation (6) indicates that the difference between the assumed character width and the width including the character candidates i and j is smaller than the threshold. In the present invention, the width direction integration processing is performed when both of the expressions (5) and (6) are satisfied.
It is performed according to Figure 29. For example, FIG. 30 (a), (b)
If the conditions of the formulas (5) and (6) are satisfied when the character candidates i, j, k are separated in the width direction as in the above, the character candidates are integrated. In the case of FIGS. 30 (a) and 30 (b), the conditions for integration are min {f (i, j), f (i, k)} = f
If (i, j) and f (i, j) <threshold, character candidate i, j
To integrate. Also, min {f (i, j), f (i, k)} = f
If (i, k) and f (i, k) <threshold, character candidates i, j,
Integrate k. Where f (i, j) is the assumed character width And the width of the circumscribed rectangle including the character candidates i and j,
f (i, k) is the assumed character width And the width of the circumscribed rectangle including the character candidates i and j.
Incidentally, the state of integration in the width direction is summarized in FIG.
(E). In the figure, f _i is the assumed character width And the width of the character candidate i. FIG. 31 (b),
As can be seen from (e), the character candidate j is the assumed character width. Even if it is located at a position exceeding
If (i, j) is smaller than the threshold, it can be integrated with the character candidate i. For this reason, for example, even a handwritten character having a large fluctuation in character width, which cannot be integrated conventionally, can be easily integrated.

In the process of the above-described character cutout processing (FIGS. 17 and 18), each time a character candidate is determined as a character example, a processed mark is added and the number of objects to be processed is reduced. Also, in each processing of FIG. 18, the horizontal direction has been described, but the vertical direction has been described.
The x and y directions are reversed. The oblique direction assumes a predetermined tilt angle, the coordinate transformation process in the angle is, the character string extraction process, included in the character cutout process (step S _2, S ₃ of FIG. 18).

Character recognition processing (step S ₃ of FIG. 2) is performed as follows. The character recognizing unit 6 refers to the character extracted by the character extracting unit 5 and a dictionary in which the character is registered in advance,
Compare and recognize one character at a time.

H. Effects of the Invention As described above, according to the present invention, an outline vectorization processing unit, a registration unit, a character string candidate extraction unit, a character string extraction unit,
Since a character cutout is provided, when reading and recognizing characters from documents and drawings containing mixed characters, line segments, and symbols, characters of any size and format can be used without restricting the size and format. Can also be read easily and accurately. In particular, since the character cutout unit is provided with the alphanumeric character recognition function and the character string type determination function, it is possible to accurately determine whether or not the character string is an alphanumeric character string. For this reason, for example, even a character that is extremely confusing when handwritten, such as the English character “R”, such as the Chinese character “Shaku”, can be accurately recognized. Also, accidental merging of alphanumeric characters can be avoided. Further, since the character extracting section is provided with the width direction integration function, even a handwritten character having a large fluctuation in character width can be easily integrated.

In addition, since character candidate data is registered in a tree structure, processing can be speeded up.

[Brief description of the drawings]

1 to 31 show an embodiment of the present invention, FIG. 1 is a block diagram of the entire configuration, FIG.
FIG. 3 is a flowchart of the contour vectorization processing unit, and FIG.
FIG. 5 is a block diagram of the contour detection device, FIG. 5 is an explanatory diagram showing a state of raster scan, FIG. 6 is an explanatory diagram showing a contour element table, FIG. 7 is an explanatory diagram showing a contour management table, and FIG. FIG. 9 is an explanatory diagram showing a contour connection table, FIG. 9 is an explanatory diagram showing a contour vector sequence table, and FIG.
(B) is an explanatory diagram showing the relationship between the pixel data and the contour element, FIG. 11 is an explanatory diagram showing the connection state between the contour elements, and FIG.
FIG. 12 is an explanatory diagram showing the direction of the contour element, FIG. 13 is an explanatory diagram of the contour vectorizing process, FIG. 14 is a flowchart of the short vector removing means, FIG. 15 is an explanatory diagram of a circumscribed rectangle, FIG.
FIG. 16 is an explanatory diagram showing a circumscribed rectangle information table, FIGS. 17 and 18 are flowcharts of character extraction processing, FIG. 19 is a flowchart of character string candidate extraction processing, and FIG.
20 (a) and (b) are explanatory diagrams of a character string candidate extraction process, FIG. 21 is an explanatory diagram of a character string extraction process, FIG. 22 is a flowchart of a character cutout process, and FIGS. b),
(C) is an explanatory diagram of the height direction separated character integration process, FIG. 24 is a flowchart of the height direction separated character integration process, FIG. 25 is an explanatory diagram of character string data after the height direction integration process, and FIG. Is an explanatory diagram showing a recognition result data table, FIG. 27 is an explanatory diagram showing a threshold value table, FIG. 28 is a flowchart of a character string type determination process, FIG. 29 is a flowchart of a width direction separated character integration process, FIG. (A) and (b) are explanatory diagrams of the width direction separated character integration process, and FIGS. 31 (a), (b), (C),
(D) and (e) are explanatory diagrams showing the state of integration of width-separated characters for each case, FIG. 32 (a) is an explanatory diagram of a character cutout error, and FIG. 32 (b) is a diagram of a character recognition unit. FIG. 1 ... Contour vectorization processing unit, 2 ... Registration unit, 3 ... Character string candidate extraction unit, 4 ... Character string extraction unit, 5 ... Character cutout unit, 6 ... Character recognition unit, 11 ... Contour Extractor, 12 ……
Outline analysis unit 13, table storage unit 31, macroscopic processing unit 32, microscopic processing unit 33, outline vector sequence table 34, short vector removal unit

Claims (1)

(57) [Claims] 1. Character data is extracted from binarized pixel data corresponding to black and white obtained by raster-scanning a subject in which characters, line segments, and symbols are mixed, and the extracted character data and a character dictionary are referred to. In a drawing reading apparatus for comparing and recognizing characters, a vector connecting two adjacent black pixels is combined based on binarized pixel data corresponding to black and white obtained by raster scanning the subject. A contour vectorization processing unit that determines data of a contour vector composed of the following and obtains as a character candidate based on circumscribed rectangle data circumscribing the contour vector, and circumscribed rectangle data of the character candidate obtained by the contour vectorization processing unit A registration unit to be registered in a tree structure, and a search range is determined from a nucleus of a character string candidate set to a predetermined size. A character string candidate extraction unit that extracts a character string Chi center coordinates are present within the search range as a candidate character string from among the data extracted by the character string candidate extraction unit, a character string candidate nuclear height h a character string extracting unit that extracts, as a character string, data in which a distance d between _c and a character candidate adjacent to each other in the character string candidate is d <h _c × k (k is a constant); When the horizontal projections of multiple adjacent character candidates in the character string extracted in
A height direction integration function that integrates these multiple character candidates,
Each character candidate integrated by the function is compared with a dictionary of alphanumeric characters and symbols to perform character recognition, and in accordance with the plurality of recognition results, the degree of approximation between the character candidate and a character in the dictionary is determined. An alphanumeric character recognition function for respectively obtaining evaluation values to be represented, and the degree of approximation between a plurality of samples of the same character and symbol and the dictionary characters and symbols corresponding to the characters and symbols of the dictionary of alphanumeric characters and symbols, respectively. Is set as the threshold value, and the set threshold value is compared with the evaluation value obtained by the alphanumeric character recognition function to determine that the character string is an alphanumeric character string. Character string type determination function to determine whether or not, the width of the character candidate determined that the determination function is not an alphanumeric character string, the assumed character width of the width equal to the size of h _c × k Is the width of the character candidate, the width of a nearby character candidate and the character Greater than the second difference between the size and the assumed character width the combined distance between neighboring character candidate,
A character cutout unit that has a width direction integration function of integrating the character candidate and a character candidate in the vicinity thereof when the second deviation is smaller than a predetermined threshold value, and that cuts out a character from a character string; A drawing reading apparatus for performing character recognition based on character data cut out by a character cutout unit.