JP2983448B2 - Drawing recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for automatically recognizing handwritten or printed line drawings with high accuracy by a computer. In particular, the present invention relates to a drawing recognition method suitable for automatically recognizing a design drawing handwritten by hand on a design sheet on which a grid, which is a grid-like writing guide line, is printed in a dropout color or the like.

[0002]

2. Description of the Related Art In a conventional method for recognizing a handwritten drawing, an input drawing is converted into a core line by thinning processing or the like, and the direction, shape, connection relationship, etc., of the line are determined based on this, and the drawing is recognized based on these. Have been.

[0003]

However, in this type of drawing recognition method, (1) line thickness information cannot be obtained.

(2) Unnecessary line segments such as whiskers are generated by the thinning process.

[0005] (3) Slight blurring of lines causes breaks and minute closed loops.

There have been problems such as the following.

An object of the present invention is to solve such problems of the conventional drawing recognition method. Since the above problems are basically caused by applying the thinning processing, the present invention provides a new drawing recognition method that is less affected by handwriting variation by a new method that does not use the thinning processing. I do.

[0008]

In order to achieve the above object, according to the present invention, first, a grid detection mark and a drop
Inputting a handwritten drawing on which the out-color grid lines are printed from the image input means ;
Grid detection mark is detected from the drawn drawing
Obtain the position of the line, then the search of the inputted drawings
Up to three handwritten lines along the grid lines
Scanning with a scanning mask of a size sufficient to capture within, and then determining a feature amount for determination for each viewing region using the region in the scanning mask as a viewing region; Is automatically converted by a threshold value to automatically determine the number of lines written on the grid lines and the thickness of each line.
The drawing recognition method is characterized in that the determined line segment is a line segment on the grid line .

In the above description, the feature amount is determined by the threshold value.
When the code conversion is performed and the number of lines written on the grid lines and the thickness of each line are automatically determined, the number of lines or the thickness of each line is corrected by majority processing with a peripheral visual field. It is preferred that

In the above, the run length of the projection result in the direction perpendicular to the grid line is used as the feature amount for determination.
Scan value and the run length value in the direction parallel to the projection result, to use one or more of the run-length value in the diagonal direction of the projection result is suitable.

[0011]

When a drawing is handwritten by hand, grid paper or the like on which a grid (grid-like writing guidelines) is printed has conventionally been used as design paper. This is because many of the written lines are written according to guidelines. That is, in many cases, a line written as a drawing is written not on an arbitrary position but on a grid.
The drawing recognition method of the present invention solves the above-mentioned problem by positively utilizing this fact.

That is, in the drawing recognition method of the present invention, the visual field is moved by moving the scanning mask along the grid, and the number of lines on the grid and the number of lines on the grid are calculated based on the characteristic amount obtained for each visual field. The thickness of each line is determined, and the determined line segment is set as a line segment on the grid. As described above, by performing drawing recognition without using the thinning processing, it is possible to make it hard to be affected by handwriting fluctuation such as positional deviation of a written line, and to obtain line thickness information. In addition, by determining the number and thickness of the lines while observing the entire field of view, it is possible to avoid the occurrence of whiskers, breaks, closed loops, and the like that occur in the thinning processing.

[0013]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a processing flow chart showing one embodiment of the present invention, in which 11 is an input binary image, 12 is a pre-processing, 13
Is a scanning position setting process, 14 is a feature extraction process, 15 is an identification process, and 16 is a process result.

Before describing each process, a design sheet to be input will be described. FIG. 2 is a diagram showing an embodiment of the design paper of the drawing, wherein 21 is a grid line, and 22 is a grid detection mark. FIG. 2A shows an example in the case of a square lattice, and FIG.
(B) is an example in the case of a triangular lattice. FIG. 2 (a),
In both of the embodiments shown in (b), rectangular grid detection marks 22 are printed at the four corners of the design paper. These grid detection marks 22 are for transmitting the position of the grid line 21 preprinted with the dropout color and the inclination of the design paper to the computer. Note that grid lines 21
And the shape, number, arrangement, and the like of the grid detection marks 22 are arbitrary, and do not limit the scope of the present invention. The printing of the grid lines 21 may be any printing as long as it can be distinguished from the line segment in the drawing.

In the pre-processing 12, a grid detection mark 22 is detected by using an input binary image 11 of the drawing input from an image scanner or the like as an input, and the position of the input binary image 11 is determined based on the position of the detected grid detection mark 22. In addition to performing the inclination correction processing, the position of each grid line 21 printed in the dropout color is obtained.

In the scanning position setting processing 13, a preprocessing 12
The center position and the inclination of the scanning mask in accordance with the position of the grid line 21 obtained in (1) (match in the direction of grid line 21)
Information is sequentially output and given to the feature extraction processing 14. This processing is repeated until the scanning of all the grid lines 21 obtained in the preprocessing 12 is completed.

FIG. 3 shows an embodiment of the flow of the feature extraction processing 14.
Shown in The feature extraction process 14 according to the present embodiment includes a visual field region setting process 31, a multi-value feature value calculation process 32, and a binary feature value calculation process 33.

First, in a visual field region setting process 31, an image of a region (visual field region) in the scanning mask is read based on the center position and inclination of the scanning mask given in the scanning position setting process 13. FIG. 4 shows an embodiment of the scanning mask. The shape of the scanning mask is square (FIG. 4A), rectangular (FIG. 4B), circular (FIG. 4C), and elliptical (FIG. 4).
(D)) is also conceivable.

In the multi-valued feature amount calculation process 32, a projected feature obtained by integrating the number of black points in the visual field in each inclination direction is extracted.
As an example of the multi-value feature, FIG. 5 shows an example in which projected features for four types of inclinations are obtained using a rectangular scanning mask. In FIG. 5, reference numeral 51 denotes a projection feature F ^{h in} the grid direction, and 52 denotes a projection feature F ^{v in the} direction perpendicular to the grid.
3 is a projection feature F ^r in the diagonally upper right direction, 54 is a projection feature F ^{l in the} diagonally upper left direction, 55 is a locally enlarged view area,
Reference numeral 56 denotes a grid line, and 57 denotes a handwritten line (an example of a double line composed of a thick line and a thin line). For example, the thickness and the number of lines parallel to the grid can be observed from the projection feature 51 in the grid direction, and in the case of the embodiment, two thin lines and thick lines can be observed from this local field of view. Further, from the projection feature 52 in the direction perpendicular to the grid, useful information for determining whether a line is broken or a dotted line / dashed line is obtained. However, the projection feature 51 alone is not sufficient to determine whether there is a break in the line as in the example of FIG. 5 or to determine a broken line or a dotted line. Therefore, using the projection feature 52 in the direction perpendicular to the grid, the presence or absence of a line break and the determination of a broken line / dotted line are determined based on the number of concave shapes obtained from the projection feature 52 and their intervals. That is, when a graphic element to be handwritten is described by a vertical or horizontal line segment as in the example shown in FIG. 5, if there are these two types of feature amounts, the attribute can be determined.

On the other hand, in the present embodiment, clear information cannot be obtained from the oblique projection features 53 and 54, but it can be used for more complicated graphic element determination such as observing line segments intersecting in the oblique direction. . However, when recognizing a drawing based on a triangular lattice as shown in FIG. 2B, the projection features 51 and 52 alone are not sufficient for recognizing a graphic element including a diagonal line segment. Yes, oblique projection features 53 and 54 are also required. For example, if a line segment on a triangular lattice uses projection characteristics in the grid direction (60 ° diagonal direction) and in a direction perpendicular to the grid direction, attribute determination can be performed in the same manner as in a normal square lattice. That is, if the projection features in the vertical direction, the horizontal direction, and the oblique direction (determined by the object) are used, it is possible to determine the attribute of a line segment in an arbitrary direction.

FIGS. 6 (a), 6 (b), 6 (c), 6 (d)
FIG. 5 shows an embodiment of a binary feature amount obtained by binarizing a projected feature with an appropriate threshold value.
The projection features F ⁱ , 60 described in the above are threshold values t ⁱ , 61 to 64
Is a binary feature quantity B ^i. Here, i indicates the direction of projection, and i {h, v, r, l}. The binary feature value B ⁱ in the direction i is obtained by inputting the projection feature F ⁱ as an input.

[0023]

(Equation 1)

Is obtained by Here, t ⁱ is a threshold value for the direction i.

FIG. 7 is an explanatory diagram of the identification processing 15.
71 is a binary feature amount, 72 is a run-length feature calculation process,
73 is a run-length feature, 74 is a code conversion process, 75
Is the identification result. In the identification process 15, first, a run-length feature calculation process 72 is performed using the binary feature value 71 as an input, and a run-length feature 73 is obtained. Next, the run-length feature is converted into a recognition result 75 by a code conversion process 74.

The case of the binary feature values 61 (FIG. 6A) and 62 (FIG. 6B) of the embodiment shown in FIG. 6 will be specifically described below. In the run-length feature calculation processing 72, a set of binary feature black run lengths, that is, run-length feature quantities (b ₁ , b ₂ ,..., B _K ) is obtained (where K is the maximum number of black runs). is there). (B ₁ , b ₂ , 0, 0,...) For the binary feature 61, (b ₁ ′, 0, 0, 0,...) For the binary feature 62 .

Various methods are conceivable for the code conversion process 74 depending on the number of attributes to be recognized. Here, an example will be described in which the maximum number (K) of line segments written on the grid is three and the line thickness is two types (thick / thin). For the black run length b _k (0 <b _k ≦ b _max ), null (0
0; no line), Thin (01; thin line), Thick
(10; bold line) and Full (11; all black). This classification can be easily realized by an appropriate threshold. Full indicates that the entire mask width (b _max ) is black.

The identification result 75 for the above example is (Thick, Thin, null) = (10, 01, 00) (Full, null, null) = (11, 00, 00).

In the above description, a relatively simple projection feature is used, since it is intended to discriminate the number and thickness of line segments. However, the present invention can be extended to the recognition of complex symbols and figures by using various features used in character recognition, as long as it is written on a grid line.

In the case where the result of distinction that is apparently discontinuous from an adjacent visual field in a certain visual field due to noise or the like peculiar to freehand, a peripheral visual field including the adjacent visual field. It is effective to further reduce the influence of noise or the like by correcting the identification result such as the number of lines or the thickness of the processing region by majority processing in between.

[0031]

As is apparent from the above description, the drawing recognition method of the present invention employs a method of judging the attribute of a line segment based on a feature amount in a visual field moving along a grid. It is characterized by being less susceptible to the effects of specific displacement and noise.

In the above, when the discrimination result such as the number of line segments or the thickness of the processing area is corrected by majority processing between the peripheral viewing areas including a certain viewing area, the situation is further improved. , Noise and the like.

In the above case, when the projection value of the black point in the visual field is used as the characteristic amount for determination,
Since the drawing can be recognized by a simple projection process and a threshold process, advantages in processing cost can be obtained as compared with the conventional method.

[Brief description of the drawings]

FIG. 1 is a process flowchart showing one embodiment of the present invention.

FIGS. 2A and 2B are diagrams showing an embodiment of a design sheet.

FIG. 3 is a diagram showing an embodiment of a flow of a feature extraction process in the embodiment of the present invention.

FIGS. 4A, 4B, 4C, and 4D are diagrams showing an embodiment of a scanning mask.

FIG. 5 is a diagram illustrating an example of a multi-value feature amount.

FIGS. 6A, 6B, 6C, and 6D are diagrams illustrating an example of a binary feature amount.

FIG. 7 is an explanatory diagram of an identification process in the embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Input binary image 12 ... Preprocessing 13 ... Scan position setting processing 14 ... Feature extraction processing 15 ... Identification processing 16 ... Identification result 21 ... Grid line 22 ... Grid detection mark 31 ... Viewing area setting processing 32 ... Multi-valued feature Amount calculation process 33 ... Binary feature amount calculation process 51 ... Projection feature in the grid direction 52 ... Projection feature in the direction perpendicular to the grid 53 ... Projection feature in the oblique upper right direction 54 ... Projection feature in the oblique upper left direction 55 ... Viewing area 56 ... Grid line 57: Handwritten line 60: Threshold value 61 to 64: Binary feature amount 71: Binary feature amount 72: Run-length feature calculation process 73: Run-length feature 74: Code conversion process 75: Identification result

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06T 7/00

Claims (3)

(57) [Claims] First, a grid detection mark and a drop
A drawing handwritten on a grid line of an outer color is input from an image input unit , and then a grid detection mark is detected from the input drawing.
Then , the positions of the grid lines are determined, and then the input drawing is sufficient to capture up to three handwritten lines in the viewing area along the determined grid lines.
Scanned by minute size of the scanning mask, then, it obtains a feature amount for determination for each viewing area to area of the scanning the mask as a viewing area, then the threshold the characteristic quantity The number of lines written on the grid line and the thickness of each line are automatically determined by code conversion , and the determined line segment is regarded as a line segment on the grid line. Characteristic drawing recognition method. 2. The code conversion of the feature value by a threshold value.
Then, when automatically determining the number of lines written on the grid lines and the thickness of each line, correcting the number of lines or the thickness of each line by majority processing with a peripheral visual field. The drawing recognition method according to claim 1, wherein: 3. The feature amount for determination is a run length value of a projection result in a direction perpendicular to a grid line and a projection value in a direction parallel to a grid line.
The resulting run length value and the run length of the diagonal projection result
2. The method according to claim 1, wherein the value is one or more of the following values:
Alternatively, the drawing recognition method according to claim 2.