JP2986185B2 - Line width classification method for line figures - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a line width classification method for line figures included in image data such as maps and mechanical drawings read into a computer.

[Conventional technology]

In order for a computer to recognize and understand an image such as a map or a mechanical drawing, it is important to classify a graphic based on a line width from an image including various types of graphics. In addition, high accuracy is required for classification of line width, and high-speed processing is required due to the enormous amount of data to be handled.

Conventionally, as a method of measuring the line width of a line figure, when the figure is viewed in directions such as up, down, left, right, and oblique, the minimum length among the lengths obtained from the number of continuous pixels existing in each direction is determined by the portion Or by scanning a line figure horizontally or vertically, obtaining the horizontal or vertical line width of the figure, and its midpoint to obtain the center line dot row and the inclination of the figure. There was a method of finding the true line width.

In addition, apply a circle template to the core line dots,
A line width is obtained from the distance between two points where the edge of the line figure intersects with the circumference (line width recognition device, Japanese Patent Laid-Open No. 229180/1985), and a center line dot row is extracted to determine the range to be extracted. There is a method of extracting peripheral pixels constituting a line having a width using a distance from a central pixel or information corresponding to the distance (line width measurement method, Japanese Patent Application Laid-Open No. Sho 60-229180). .

[Problems to be solved by the invention]

However, any of the above methods actually measures the line width, is difficult to process, and cannot be said to be suitable for high-speed processing. For this reason, it has been difficult to speed up the process of classifying various types of figures by line width in order to recognize and understand images such as maps and drawings with a computer.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a line width classification method capable of classifying various figures with high accuracy by combining basic processing and enabling high-speed processing by hardware or parallel calculation.

[Means for solving the problem]

In order to solve the above-described problems, the present invention provides a convolution integrator that performs convolution integration on a load matrix with respect to image data read by a computer, and performs a process of setting a pixel value less than a predetermined threshold to 0 with respect to the image data. A threshold processing unit for performing the arithmetic operation between the pixels at the corresponding positions between the two image data, and an inter-image operation unit for performing the arithmetic operation. Classified by The convolution integral has a line width contraction load matrix in which the central portion has a positive value and the peripheral portion has a negative value, and a line width expansion load matrix in which the central portion has a positive value and the value approaches 0 as the distance from the center increases. Prepare Then, convolution integration is performed by the product-sum operation of these load matrices and image data.

That is, for the read image data, convolution integration is performed by a sum-of-products operation using a load matrix having a positive value at the center and a negative value at the periphery, and sets a pixel value less than a determined threshold to 0, The pixel value less than the threshold value determined by subtracting the image data obtained in the above from the original image data is set to 0, and the value of the image data obtained in the above process becomes 0 as the central portion becomes positive and moves away from the center. Is convolved by a product-sum operation with a load matrix approaching, and the pixel value less than a threshold determined by subtracting the image data obtained by this from the original image data is set to 0.

[Action]

The line width classification method of the present invention classifies a figure by line width by convolution of image data and a load matrix, threshold processing, and arithmetic operation between image data.

That is, first, the line width of a line figure is contracted by sequentially executing convolution integration, threshold processing, inter-image calculation, and threshold processing using a line width contraction load matrix. At this stage, the line figure with a small line width disappears. next,
The contracted line figure is expanded by convolution integral with the line width expansion load matrix. At this time, no change occurs in the line graphic having a thin line width which has already disappeared. When the image data obtained at this time is subtracted from the original image data by an inter-image operation and threshold processing is performed, only a line figure with a small line width is extracted.

〔Example〕

FIG. 1 is a system diagram for explaining an embodiment of the present invention. In FIG. 1, a drawing to be processed is read as a grayscale image by an image reading unit 8 such as an image scanner and stored in a memory 9. Then, the line data extraction processing is performed on the image data stored in the memory 9 by the two line width extraction units 10 and 11 to extract line figures having two different line widths. That is, as described later, these line width extraction units 10 and 11 perform convolution integration using a load matrix, but the line width extraction unit 10 and the line width extraction unit 11 use different weight matrices. , The border line for extracting the line figure up to which line width is different. The line figures smaller than the different line widths extracted by the two line width extraction units 10 and 11 are stored in the memories 12 and 13, respectively.

Next, in the inter-image calculation unit 14, the two memories 12, 13
Then, a line graphic having a specific line width is extracted by subtracting between the pixels at the corresponding positions from the image data composed of the line graphic smaller than the different line width stored in. For example, if the load matrix used by the line width extraction unit 10 is set to be larger than the load matrix used by the line width extraction unit 11, the output result of the line width extraction unit 10
The output result of is subtracted. As a result, only a line figure having a line width equal to or less than the border line in the line width extraction unit 10 and equal to or larger than the border line in the line width extraction unit 11 is obtained, and is output as a line width classification result.

FIG. 2 shows a procedure of a graphic extraction process based on a line width performed in the line width extracting units 10 and 11 shown in FIG. 1. FIG. 3 shows, as an example, two figures having different line widths. 3 shows the result of performing the line width classification process on. less than,
The procedure of line width classification will be described with reference to FIGS. In FIG. 3, (a) shows two line figures having different line widths drawn on the screen, and the cross section in the level direction of the original image data is (0). I do. FIGS. 3 (1) to 3 (9) also show cross sections in the level direction of image data, of which (1) to (7).
Indicates a cross section of the image data obtained in each of steps 1 to 7 in FIG. In each cross section, the horizontal axis indicates the position on the screen, and the vertical axis indicates the density of the image. In this example, the reference value is set to 0.

In FIG. 2, first, the difference (DOG) between two Gaussian functions with respect to the original image data (FIG.
Function) to perform convolution integration. The DOG function concentrically forms a load matrix having a positive value at the center and a negative value at the periphery, and is expressed by the following (Equation 1).

DOG (r) = (1 / 2πσ 1 2) · exp (-r 2 / 2σ 1 2) - (1 / 2πσ 2 2) · exp (-r 2 / 2σ 2 2) ... ( 1) where, σ ₁ , σ ₂ : standard deviation of Gauss function (σ ₂ / σ ₁ =
1.6), r: distance from the center of the concentric circle.

That is, in the convolution integral here, a matrix of size M × M smaller than the screen is created by the above (Equation 1) (M / 2 = 3σ ₂ ), and the product sum is calculated between this matrix and the original image data. Shall be taken. By sequentially performing the convolution integration processing while scanning the matrix on the screen, image data as shown in FIG. 3A is obtained.

Next, in the threshold processing 2, the pixel value having a value less than the determined threshold is set to 0 with respect to the result obtained by the convolution integral 1. Here, the threshold value is set to 0, which is the same value as the reference value, but may be a value larger than the reference value. As a result, in the case of a thick line, a portion having a pixel value of 0 is formed at the center, but in a case of a thin line, a portion having a pixel value of 0 is not formed (FIG. 3 (2)).

In the inter-image operation 3, the threshold processing 2 is performed based on the original image data.
(However, when subtraction is performed between images, a sufficiently large coefficient is applied to the image data to be subtracted in FIG. 3 (2)). The subtraction is performed between the pixels at the corresponding positions. Thereby, FIG. 3 (3)
Is obtained.

Then, in the next threshold processing 4, threshold processing is performed on the result obtained by the inter-image calculation 3 with 0. That is, 0
The pixel value of the value less than is set to 0. As a result, in the case of a thick line, both ends of the line are slightly reduced so that the line width is reduced as compared with the original image data, and in the case of a thin line, the entire line disappears (FIG. 3 (4)).

Next, in the convolution integral 5, Gauss based on the following (Equation 2) is applied to the image data obtained by the threshold processing 4 above.
Performs convolution integration with the s function to expand the contracted line figure. The Gaussian function forms a load matrix in which the central portion is a positive value and the value approaches 0 as the distance from the center increases.

Gauss (r) = (1 / 2πσ ² ) · exp− (r ² / 2σ ² ) (Equation 2) Here, the standard deviation σ of the Gauss function is σ _{2 of the} above (Equation 1).
And the size M of the matrix is also the same value. Here, there is no change in the thin line in which the line graphic has disappeared by the previous processing, but in the case of the thick line which has not disappeared, the line width is expanded to at least the line width of the original image. (FIG. 3 (5)).

Further, in the inter-image operation 6, the image data obtained by the convolution integral 5 is subtracted from the original image data (again, the image data to be subtracted is multiplied by a sufficiently large coefficient). Result (Fig. 3 (6))
On the other hand, when threshold processing is performed at 0 by the threshold processing 7, the thick line disappears and the thin line is restored (FIG. 3 (7)). According to the procedure described above, of the two figures having different line widths, the figure having the larger line width can be deleted, and only the figure having the smaller line width can be extracted.

Note that the line width of a graphic that can be deleted depends on the standard deviation of the DOG function. Accordingly, only the thin line is extracted by the line width extraction unit 11 of FIG. 1, and both the thick line and the thin line are extracted by the line width extraction unit 10 which sets a larger standard deviation value (FIG. 3 ( 8)), by subtracting the image data of FIG. 3 (7) from the image data of FIG. 3 (8) in the inter-image calculation unit 14 of FIG. 1, only the thick line can be extracted (FIG. 3 (9).

FIG. 4 shows the result of performing line width classification processing on eight types of line figures having a line width of 1 to 8 pixels. In FIG. 4, first, convolution integration is performed on the original image data (FIG. 4 (a)) using the DOG function based on the above (Equation 1), and threshold processing is performed at 0, and FIG. 4 (b)
Is obtained. This is because, as described above, in the case of a thick line, a portion having a pixel value of 0 can be at the center,
A thin line indicates that a portion having a pixel value of 0 cannot be formed.

Next, the image data generated by the above threshold processing is subtracted from the original image data by an inter-image operation, and the obtained result is thresholded at 0. The image data as shown in FIG. 4 (c) is obtained in which the line width is shrunk and the line width is shrunk.

Next, the contracted line figure is expanded by performing convolution integration on the image data of FIG. 4C using the Gauss function based on the above (Equation 2) (FIG. 4D). Further, the image data obtained by the convolution integration based on the above (Equation 2) is subtracted from the original image data by the inter-image operation, and the obtained result is subjected to a threshold processing with 0, whereby a line thicker than a certain bit width is obtained. Disappears, and image data as shown in FIG. 4 (e) in which a line thinner than a certain bit width is restored is obtained. According to the above-described procedure, a graphic having a specific line width or more can be erased, and only a graphic which is thinner than the specific line width can be extracted.

As described above, the line width of a graphic that can be erased depends on the magnitude of the standard deviation of the DOG function. Therefore, the two line width extraction units 10 and 11 shown in FIG. By extracting a graphic and performing a subtraction between the respective result images by the inter-image calculation unit 14, a graphic in a specific line width range can also be extracted.

FIG. 4 (e) shows the result of extracting a line segment having a line width of 4 bits or less. FIG. 4 (f)
Shows the result of performing convolution integration with a larger standard deviation than the case of FIG. 4 (e), that is, a large size load matrix, and extracting a line width of 6 bits or less. In this case, by subtracting the image data of FIG. 4E from the image data of FIG.
Only 6-bit line segments can be extracted and classified (FIG. 4 (g)).

In the above embodiment, a function whose value changes concentrically is used as a load matrix when performing convolution integration, but the present invention is not limited to this. For example,
If the load matrix has a positive value in the center and a negative value in the periphery, or a load matrix with a positive value in the center and the value approaches 0 as the distance from the center increases, the value follows a square or other shape May be used.

〔The invention's effect〕

As described in detail above, according to the present invention, a map,
When classifying figures on an image such as a mechanical drawing by line width, figures of any shape can be processed using a combination of basic image processing methods such as product-sum operation, threshold processing, and inter-image operation. Thus, high-speed processing can be performed by dedicated hardware or parallel processing.

[Brief description of the drawings]

FIG. 1 is a system diagram illustrating an embodiment of the present invention, FIG. 2 is an explanatory diagram of a line width extracting process, and FIG. 3 is a diagram illustrating a line width classifying process for line figures having two different line widths. FIG. 4 is a diagram showing a cross section of the pixel in the level direction as a result of the execution, and FIG. 4 is an image showing a result of performing a line width classification process on an image in which line figures having eight different line widths are drawn. In the reference numerals used in the drawings, 1 ... convolution integral 2 ... threshold processing 3 ... calculation between images 4 ... threshold processing 5 ... convolution integral 6 ... calculation between images 7 ... threshold processing 8 ... image Reading unit 9 Memory 10 Line width extraction unit 11 Line width extraction unit 12 Memory 13 Memory 14 Image calculation unit.

Continuation of the front page (56) References JP-A-53-118941 (JP, A) JP-A-50-135948 (JP, A) JP-A-60-229180 (JP, A) JP-A-62-192606 (JP) , A) (58) Field surveyed (Int. Cl. ⁶ , DB name) G06T 7/00 JICST

Claims (3)

(57) [Claims] 1. A convolution integral is performed on a read image data by a product-sum operation using a load matrix having a positive value in a central portion and a negative value in a peripheral portion, and a pixel value less than a predetermined threshold value is set to 0. A second step of subtracting the image data obtained in the first step from the original image data to set a pixel value less than a predetermined threshold to 0, and an image data obtained in the second step In contrast, convolution integration is performed by a product-sum operation with a load matrix whose value approaches 0 as the center part is a positive value and moves away from the center, and the image data obtained thereby is subtracted from the original image data, and is less than a determined threshold. And a third step of setting the pixel value of 0 to 0. 2. A convolution of the read image data by a sum-of-products operation using a load matrix consisting of a positive value at the center and a negative value at the periphery in a concentric manner, and pixel values less than a predetermined threshold value A second step of subtracting the image data obtained in the first step from the original image data and setting a pixel value smaller than a predetermined threshold to 0, and a second step of subtracting the image data obtained in the first step from the original image data. The image data obtained is subjected to convolution integration by a sum-of-products operation with a load matrix whose value approaches 0 as the center is positive and moves away from the center, and the obtained image data is subtracted from the original image data. And setting a pixel value smaller than the threshold value to 0. 3. A method according to claim 1, wherein two image data are generated by a convolution integral using weight matrices having different sizes, and a corresponding pixel of the generated two image data is generated. A line width classification method for a line figure, wherein subtraction is performed.