JPH0816931B2 - Contour extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] The present invention relates to a method of image data processing, and more particularly to a method of accurately and rapidly extracting a contour line necessary for determining an outer shape of input image data.

[Industrial applications]

With the progress of image processing technology, its application side has expanded and TV
Applications such as discriminating the outer shape from an image input by a camera, using it for handling, inspection, or reading characters are expanding. For such applications, a higher speed and easier hardware-based outline extraction method is required as a means for executing the extraction of the outline of an image in real time.

[Conventional technology]

In the image processing, as is well known, the image is decomposed into pixels, or points of a unit called PIXEL, etc., and these are accumulated corresponding to each word of a storage device having an orthogonal coordinate address, and usually a pixel unit corresponding to one word. Address and execute the process. Each pixel is usually 3 to 8 bits indicating a multi-valued density (or luminance) level, and R, G, B indicating the constituent primary colors in a color image.
It is composed of one word to which information about each of is added. Since the present invention is a technique for processing density, a black and white image will be described. Further, since the contour extraction of an image having only binary values (only white and black) as the density level is extremely easy, it is not mentioned here, and an image having a density level of several tens of values or more is taken as an example here.

Conventionally, a method of processing by software has been mainly used for extracting the contour line of an image because of the complexity of the algorithm. An example of that will be shown below, and the algorithm of the processing will be briefly described.

 FIG. 6 shows an example of the distribution of pixels in the storage device.

In the figure, the distribution of the pixels in the grid pattern is an enlarged view of a part of the image storage device.
~ X6, Y1 to Y5, and the number written on each pixel indicates the density of that pixel. In order to detect the contour of the image from the pixels distributed in this way, conventionally, a processing procedure whose flow is shown in FIG. 7 has been used. This is to obtain the density gradient vector by differentiation, and perform an arithmetic operation between each pixel and its neighboring pixels according to its direction and size,
This is a method of determining a contour line by detecting a pixel address having the maximum density gradient along the direction of the density gradient.

Here, the image differentiation will be described with reference to FIG. Assuming that the point for which the differential coefficient is obtained is P, attention is paid to eight pixels 00 to 22 adjacent to the point P on the diagram, the densities of the respective points are indicated by the symbols L _{00 to} L ₂₂ , and X and Y at the point P are shown. The differential coefficient with respect to the direction is defined as Dx and Dy, and is defined as follows.

_{Dx = (L 22 + L 12} + L 02 -L 20 -L 10 -L 00) / 3 Dy = (L 22 + L 21 + L 20 -L 02 -L 01 -L 00) / 3 for 6 pixels This operation adjacent This is equivalent to obtaining the differential coefficient using the average value of, and is performed in the differential operation section (each section in the X and Y directions) shown in FIG.

Next, using the resulting values of Dx and Dy, Fp = (Dx ² + Dy ² ) ^1/2 is obtained as the absolute value of the concentration gradient (hereinafter abbreviated as Fp) in the concentration gradient calculation unit. Further, in the gradient direction calculation unit, Gp = COS ⁻¹ (Dx / Fp) (Dy ≧ 0) Gp = −COS ⁻ for pixels whose Fp is greater than a predetermined value as the direction of the density gradient (hereinafter abbreviated as Gp) Calculate ¹ (Dx / Fp) (Dy <0). Further, when the magnitude of the density gradient is less than the predetermined value, a special value such as 999 is output.

The above calculation is sequentially performed for each pixel on the image storage device shown in FIG. 6, and as a result, Fp and Gp are obtained for each pixel.

Next, the following algorithm is used to determine the contour in the contour extraction unit based on Fp and Gp obtained for each pixel.

As shown in FIG. 9, the circumference is equally divided into eight sectors, each of which is labeled with a to d. The range of each sector is set as follows, where the positive direction of the X axis is 0.

Sector a: π / 8> Gp ≧ −π / 8 and the direction where this is rotated π length Sector b: 3π / 8> Gp ≧ π / 8 and the direction where this is rotated π length Sector c: 5π / 8> Gp ≧ 3π / 8 and the direction in which it is rotated π length Sector d: 7π / 8> Gp ≧ 5π / 8 and the direction in which this is rotated π length Depending on which sector Gp is included, the following conditions It is determined that the pixels satisfying the above are pixels forming the contour.

 When Gp is included in sector a, Fp (x, y) ≧ Fp (x−1, y) and Fp (x, y) ≧ Fp (x + 1, y).

 When Gp is included in sector b, Fp (x, y) ≧ Fp (x + 1, y + 1), and Fp (x, y) ≧ Fp (x−1, y−1).

 When Gp is included in sector c, Fp (x, y) ≧ Fp (x, y + 1) and Fp (x, y) ≧ Fp (x, y−1).

 When Gp is included in sector d, Fp (x, y) ≧ Fp (x + 1, y−1), and Fp (x, y) ≧ Fp (x−1, y + 1).

The above operation divides the direction of Gp into eight directions of π / 4, compares the pixel P shown in FIG. 8 with each pixel of 00 to 22 corresponding to the eight directions, and finds that Fp is maximized in the direction of Gp. This is an operation to extract the pixels that have become. For example, in FIG. 5, if Gp obtained for the pixel at the position of X3, Y3 is in the direction of the arrow in the figure, it is determined from the above logic that this pixel is a pixel forming a contour line.

In this way, the pixels forming the contour are extracted,
In particular, the contour extraction operation is usually performed by program control because of its complexity.

[Problems to be solved by the invention]

As described above, in the conventional image processing method,
In order to extract the contour line, it is necessary to perform the calculation according to the algorithm described in the previous section for all the pixels forming the image, which is usually executed by program control. For this reason, the processing time becomes remarkably long, and in order to extract the contour in real time, hardware and a means for speeding up this are required.

[Means for solving problems]

The present invention uses a contour line extraction method whose principle is shown in FIG. 1 in order to solve the above problems.

The density gradient vector calculation unit 2 obtains and outputs Fp3 and Gp4 at each image address from the image information input from the pixel input unit 1.

When Gp4 at this address is within a range of a positive / negative predetermined angle (for example, ± 45 °) with respect to the positive / negative direction of the X axis, the X-direction concentration gradient extraction unit 5X determines a positive / negative predetermined angle with respect to the positive / negative direction of the Y axis (for example, When it is within the range of ± 45 °, the Y-direction density gradient extraction unit 5Y outputs Fp3. When the above conditions are not met, the numerical value "0" is output.

Maximum value filter 6 X, 6Y is the concentration gradient extraction unit X, Y direction calculation unit 5
Regarding the Fp of each pixel of the X, 5Y outputs, the maximum value in the Fp of the pixels contained in the addresses before and after a certain number of constants in the X and Y axis directions is substituted and output.

The outputs of the maximum value filters 6X and 6Y are compared with the Fp3 of the pixel at the same address in the comparison operation units 7X and 7Y. If they match, the pixel address is given a logical "1", and if they are not the same, a logical "0" is given. Output.

The logical sum unit 8 creates the logical sum of the outputs of the two systems of the comparison operation units 7X and 7Y, and this is obtained as the contour line output 9.

[Action]

To extract the pixel with the maximum differential value from the density differential vector as the contour line, the direction of the differential vector is X,
The image is classified into two groups close to each direction of Y, the image processed by the maximum value filter is compared with the original image, and the two groups are again combined by the logical sum to extract the contour line.

〔Example〕

FIG. 2 shows an embodiment of the contour line extraction method according to the present invention. The image input unit 1 and the X- and Y-direction differential calculation unit 21 in this figure
The operations of X, 21Y, their outputs Dx, Dy, the concentration gradient calculation unit 22, and the gradient direction calculation unit 23 are the same as those described in the conventional example, and thus the description thereof is omitted.

Processing the image containing the circular object shown in FIG. 3A
Fp and Gp are obtained, and as a result, the third
It is assumed that the distributions shown in Figures B and C are obtained. Considering the thickness of the line in Fig. 3B as the value of Fp, naturally at the location of the contour
The value of Fp is large. Further, in FIG. 3C, the direction of Gp is indicated by an arrow, but it is clear that the original image is circular and therefore radially distributed from the center. In the figure, Gp is conceptually shown only for the contour portion.

Gp is first input to the threshold processing units 51X and 51Y and is classified into a direction close to the X axis and a direction close to the Y axis. That is, the threshold value processing unit 51X is in the range of ± 45 ° with respect to the positive / negative direction of the X axis, and the threshold value processing unit 51X is in the range of ± 45 ° with respect to the positive / negative direction of the Y axis.
Y outputs “1”.

Here, the outputs of the threshold value processing units 51X and 51Y are as shown in FIGS. 3D and 3E.

One of the outputs of the threshold value processing units 51X and 51Y indicates a logical "1", and when this is input to the product calculation units 52X and 52Y and the product with Fp is taken, as a result, Fp is classified according to the direction of Gp. It
Fp in the area where the output of the threshold processing unit 51X is "1"
Is output from the product calculation unit 52X (FIG. 3F), and the threshold value processing unit 5
Fp in the area where the output of 1Y is “1” is the product operation unit 52
It is output from Y (Fig. 3G).

The significance of the threshold value processing unit is that, in order to effectively execute the maximum value filtering to be performed next, it is necessary to select pixels having Gp in a direction close to parallel to each X, Y axis direction to be processed. It is in. Further, the processing along the two axes of X and Y is also because the address of the image storage device is the X and Y address, so scanning is simple and hardware implementation is easy.

The operation of the maximum value filter 6X will be described with reference to FIG. The upper part (A) of the figure shows the distribution of Fp on the storage device before the processing. Here, it is assumed that the X-axis direction is distributed as shown in the figure. Set "Window" in the X-axis direction as shown in the figure for filter calculation. This is an address continuous about 5 to 9 in the X-axis direction around the pixel address to be processed, and the pixels in this window are sequentially read out, and the maximum Fp value among them is output. In the window shown in the figure, the maximum value of Fp in the window is 5, so 5 is written in the center address as shown by the dotted line. This is shown as the distribution of pixels after processing in the lower part (B) of the figure. Next, shift the window by one address and move to the window in the figure, and similarly enter the maximum value Fp in the address in (B). Thus, in the figure, the processes up to are shown. As a result of the maximum value filtering, the contents of (A) are changed to (B) by the filter calculation.

This operation is exactly the same as that of the maximum value filter 6X except that the window is set in the Y axis direction for the maximum value filter 6Y and the processing is advanced in the Y axis direction. As a result, before processing
Although the number of pixels showing the maximum value of Fp was one, as a result of the processing, it is spread to five widths, and it is easy to understand that this width is equal to the width of the window. Fig. 3 H and I have maximum value filter 6
The pixel distribution of the X and 6Y outputs is shown. In the figure, a black line with a width is a region in which the maximum value of Fp is distributed, and other parts conceptually show that the value of Fp is different from the original value.

The significance of this operation is to extract the point where Fp takes the maximum value in order to determine the position of the contour line, instead of repeating comparison and determination with Fp of pixels in the vicinity of it as in the conventional example, instead of repeating it. This is a preprocessing for extracting the point where Fp has the maximum value only by the comparison operation described below. Therefore, it is performed for the purpose of expanding and emphasizing the distribution of the maximum value of Fp.

As shown in FIG. 5, the comparison operation units 7X and 7Y compare the values of Fp before and after the processing by the maximum value filter pixel by pixel, and write a logical value “1” in the portion having the same value. This is an operation for extracting the contour line. This is emphasized by expanding the width of the distribution while the maximum value of Fp remains unchanged from the maximum value filtering as is clear from FIG. 5, and the Fp value changes in the parts other than the maximum value. By simple comparison, contour lines are extracted as shown in FIGS.

Since the processing between the threshold value processing units 51X and 51Y and the comparison operation units 7X and 7Y is executed for each direction of Gp, and each pixel is processed in one of the columns X and Y, this is finally performed. Are added by a logical sum and put together to complete a contour line output as shown in FIG. 3L.

Comparing the above operation with the conventional example, detection of the point where Fp has the maximum value was conventionally performed by calculating the magnitude relationship between the pixel and Fp of the adjacent pixel, but in the method of the present invention, the maximum value is calculated. By using the filter, it is possible to execute it by repeating a simple operation in the form of scanning the image storage device. Further, the processing between the threshold value processing units 51X and 51Y between the comparison operation units 7X and 7Y is different only in the X and Y axes, and the other processes are exactly the same.
The same circuit can be used. This is a method suitable for hardware implementation, and particularly when a pipeline method or the like is applied, speedup can be easily realized.

〔The invention's effect〕

By implementing the present invention, it becomes possible to extract a contour line by a simple algorithm, which facilitates hardware implementation and accompanying speeding up.

[Brief description of drawings]

FIG. 1 is a principle diagram of the contour line extraction method according to the present invention, FIG. 2 is an embodiment of the contour line extraction method according to the present invention, and FIG. 3 (A to L) are specific examples of the processing according to the present invention. FIG. 4 is a description of the maximum value filter, FIG. 5 is a description of comparison operation, FIG. 6 is a pixel distribution in the image storage device, FIG. 7 is a flow of contour extraction by a conventional method, and FIG. 8 is image differentiation. Example Method, FIG. 9 shows an example of contour extraction. In FIGS. 1 and 2, 1 is an image input unit, 2 is a density gradient vector calculation unit, 21X, Y
Is an X, Y direction differential calculation unit, 22 is a concentration gradient calculation unit, 23 is a gradient direction calculation unit, 3 is a concentration gradient, 4 is a gradient direction, and 5X, Y is X, Y.
Directional concentration gradient extraction unit, 51X, Y is a threshold processing unit X, Y, 52X, Y is a product operation unit X, Y, 6X, Y is a maximum value filter X, Y, 7X, Y is a comparison operation unit X, Y, Reference numeral 8 is a logical sum unit, and 9 is a contour line output.

Claims (1)

[Claims] 1. Addressing with an orthogonal axis address,
An image processing apparatus for processing an image composed of a set of pixels having multi-valued density levels to extract contour lines of the image, processing pixels input from an image input unit (1), A density gradient vector calculation unit (2) for outputting the density gradient (3) and the gradient direction (4) in the pixel, and the gradient direction (4) for a pixel having a direction within a predetermined angle between the positive and negative directions of the X axis. The concentration gradient (3)
For a pixel having a direction other than the above range, an X-direction density gradient extraction unit (5X) that outputs a numerical value “0”, and the gradient direction (4) has a direction within a predetermined angle between the positive and negative directions of the Y axis. For the pixel, the density gradient (3)
For pixels having directions other than the above range, a Y-direction density gradient extraction unit (5Y) that outputs the numerical value “0”, and for the output of the X-direction density gradient extraction unit (5X), the X-axis direction and the Y-direction density Y for the output of the gradient extractor (5Y)
A maximum value filter X, Y (6X, 6Y) for selectively outputting the maximum value of the density gradient (3) in the pixels included in a predetermined neighborhood section centered on the input pixel address in the axial direction; A comparison operation unit X, Y (7X, 7Y) that compares the output of the maximum value filter X, Y (6X, 6Y) with the density gradient (3) at the same address and outputs a logical value "1" only when they match. ) And a pair of comparison operation units X, Y (7X, 7Y) output as a contour line output (9).