JP3536589B2 - High-speed Huff converter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a Hough transform used as a representative method for extracting a straight line from an input image in fields such as scene recognition and character recognition.

[0002]

2. Description of the Related Art Hough transform is performed by, for example,
E-Transaction Pattern Analysis 11 (1
989) pp. 429-439 (FH. CHENG et a
l. "Recognition of handwritten Chinese characters
by modified Hough transformtechniques. ", IEEE Tran
s.Pattern Anal.Machine Intell, vol.11 (1989) P.42
As described in 9-439), it is used for the recognition of straight lines that compose a shape as one method of feature extraction when recognizing characters and figures.

[0003] The principle of the general Hough transform will be described. In the Hough transform, a straight line is applied to a set of discrete pixels, utilizing that a straight line is uniquely determined by an angle θ between a distance ρ from the origin and a coordinate axis. That is, the calculation of (Equation 1) is performed on the pixels constituting the image (hereinafter, referred to as black pixels), and ρ which can take a straight line passing through the pixels is obtained.
And all combinations of θ.

[0004]

(Equation 1)

This combination draws a curve called a Hough curve on a Hough space represented by ρ and θ. When the Hough curves for all the black pixels on the image are obtained, the Huff curves obtained from the pixels constituting the straight line intersect at one point, so the straight line that is applied to a set of discrete pixels from the combination of ρ and θ at the intersection of the Hough curves Can be uniquely determined. When Hough transform is performed on an actual image, a large number of Hough curves are obtained, and there are many intersections between the Hough curves. For this reason, a specific threshold is usually set, and a straight line is obtained from a combination of ρ and θ at a point where the number of Hough curves equal to or greater than the threshold intersects. FIG. 2 shows the correspondence between pixels in the image space and Hough curves in the Hough space. In FIG. 2, a pixel A in the image space corresponds to a Hough curve a in the Hough space. Similarly, pixel B corresponds to Hough curve b, and pixel C corresponds to Hough curve c. Intersection of the Huff curves a to c (θ
c, ρc) corresponds to the angle between the coordinate axis of the straight line passing through the pixels A to C and the distance from the image origin.

[0006] In the conventional Hough transform apparatus, while changing θ from 0 to π for one pixel, the operation of (Equation 1) is performed to obtain ρ. The calculation amount when the angular resolution is 1 degree is the calculation amount per pixel (Equation 1) * 180 times.
Next, the memory corresponding to the Huff space (Huff space memory)
Above, for all the combinations (θ, ρ) of ρ and θ constituting the Hough curve, the data of the corresponding address is incremented by one, and the degree of overlap of the Hough curve is stored. Such processing is performed for all black pixels present in the processing target image, and a straight line is obtained from the Huff space memory from a combination of ρ and θ having a large degree of overlap of the Hough curves.

[0007]

The Hough transform process is used when extracting a straight line from an image input in fields such as character recognition and robot vision, and requires high-speed processing. However, it is necessary to perform two multiplications and one addition of the trigonometric function on all the black pixels existing in the image while changing the angle parameter θ from 0 to π. For this reason, when the image size increases, the amount of calculation required for processing increases dramatically.

SUMMARY OF THE INVENTION It is an object of the present invention to reduce the amount of calculation required for conversion between a Huff transform operation target and one operation target, and to speed up processing.

[0009]

According to a first aspect of the present invention, there is provided a high-speed Hough transform apparatus including a pixel to be processed and an image constituting the image to be included in a rectangle based on a specific rule, and having a vertex coordinate of the rectangle. Inclusion processing means for outputting, Huff operation means for performing a Huff transform operation with the vertex coordinates of the rectangle as an input, and outputting the operation result, and Huff curve in the Hough space with the operation result of the Hough transform for the rectangle vertex coordinates and the rectangle as the inputs And a Huff space memory for recording the degree of overlap of

According to a second aspect of the present invention, in the high-speed Hough transform apparatus, the inclusion processing means includes all pixels constituting the image with the pixels constituting the image not included in the rectangle as base points. If the ratio of the number of pixels constituting the image to the number of other pixels is larger than a specific threshold value, one or both of the coordinate axes in the image space to track and include the pixels constituting the adjacent image The length of the side of the rectangle is extended, and when the ratio of the number of pixels constituting the image to the number of other pixels is equal to or less than the threshold value, and when the object is not tracked and is not included in another rectangle, The rectangle is generated so that the length of the side is determined when there are no more pixels constituting the image, the coordinates of the vertices are output, and the rectangle having the number of pixels equal to or smaller than a threshold is removed.

According to a third aspect of the present invention, in the high-speed Hough transform apparatus, the Hough calculating means forms the coordinate axis with a straight line obtained by extending the diagonal line of the rectangle with the vertex coordinates output by the inclusion processing means as an input. The angle is calculated as the calculation angle of the Hough transform, and the angle between the coordinate axis passing through the center point of the rectangle and the coordinate axis is the calculated angle, and the straight line passing through the center point of the rectangle and the coordinate axis is 0 ° and The distance from the origin of the image space is calculated for a straight line having an angle of 90 degrees with the coordinate axis passing through the center point of the rectangle,
The calculation angle and the calculation result used in the calculation are output.

According to a fourth aspect of the present invention, in the high-speed Hough transform device, the Hough space memory receives the vertex coordinates and the operation result as input and stores the vertex coordinates and data of an address corresponding to the operation result into the vertex coordinate. Is added to the result of the specific operation.

[0013]

Embodiments of the present invention will be described below. FIG. 1 shows a configuration of a Hough transform apparatus according to one embodiment of the present invention. In FIG. 1, reference numeral 1 denotes inclusion processing means, 2 denotes Huff operation means, and 3 denotes Huff space memory.

The inclusion processing means 1 receives a processing target image as an input, includes a set of black pixels satisfying a specific condition in a rectangular area, and outputs the vertex coordinates of the rectangle. The Huff calculation means 2 receives the coordinates of the vertices of the rectangle output from the inclusion processing means 1 as input, performs a Hough transform calculation, and outputs a calculation angle θ and a calculation result ρ. The Huff space memory 3 receives the coordinates of the vertices of the rectangle output from the inclusion processing means 1, the calculation angle θ and the calculation result ρ output from the Hough calculation means 2, and increments the data of the corresponding address by incrementing the Hough curve. Is stored.

Hereinafter, a more specific operation of the high-speed Hough converter of the present invention will be described. When an image to be processed is input, the inclusion processing means 1 performs a raster scan on the input image, and when a pixel that is not yet included in the rectangle 4 among the black pixels 4a is reached, the rectangle 4 is generated with that pixel as a starting point. Start. When generating the rectangle 4, as shown in FIG. 3, it is checked whether the ratio of the number of black pixels 4a included in the rectangle 4 to the number of other pixels (hereinafter, referred to as white pixels) 4b is larger than a threshold value. However, the side of the rectangle is extended in parallel with the coordinate axis of the image space so as to include the adjacent black pixel 4a. When the ratio of the number of the black pixels 4a and the number of the white pixels 4b included in the rectangle becomes equal to or less than the threshold value, when there is no adjacent black pixel 4a, or when all the adjacent black pixels 4a
If a is already included in another rectangle 4, the coordinates of the vertices of the rectangle 4 are output, and a raster scan is performed to search for a pixel to be the starting point of the next rectangle generation. When the size of the rectangle 4 is equal to or smaller than the threshold value, the vertex coordinates of the rectangle 4 are not output, and are excluded from the target of the Hough transform processing. Such processing is performed until all the black pixels 4a on the processing target image are included in the rectangle.

The Hough operation means 2 connects the coordinates (xc, yc) of the center point 5 of the rectangle 4 to the apex angle based on the vertex coordinates of the rectangle 4 output by the inclusion processing means 1, as shown in FIG. The angles θ1 and θ2 between the straight lines 6 and 7 and the coordinate axes are obtained. Next, according to (Equation 1), the angles passing through the point (xc, yc) and the coordinate axis are θ1, θ2, 0, and 90.
The distance between the straight lines 6 to 9 and the origin of the image is output. The Huff space memory calculates the ratio between the length of the long side and the short side of the rectangle from the coordinates of the vertices of the rectangle, and calculates the calculation angle output from the Huff calculation means 2 and the contents of the address on the Huff space memory corresponding to the calculation result. Is added to.

The threshold value for determining the ratio of the number of black pixels to the number of white pixels contained in the rectangle used to determine the size of the rectangle depends on the type of the image to be processed, but at least the number of black pixels and the number of white pixels A value may be set such that the number of black pixels increases with the lower limit of the case where the ratio of numbers is 6: 4. The threshold value for determining the number of pixels of the rectangle to be removed depends on the type of the processing target image, but may be set to a value about the number of pixels constituting noise included in the processing target image.

It is not necessary to wait until the processing by the inclusion processing means 1 is completed in the processing by the Hough operation means 2, and the processing can be performed at any time when the vertex coordinates of the rectangle 4 are output from the inclusion processing means 1. It is.

As described above, all the black pixels constituting the image to be processed are included in a rectangle, and the rectangle is subjected to the Hough transform. Therefore, the number of the Hough transform can be reduced. Furthermore, by excluding a rectangle having the number of pixels equal to or less than a specific threshold from the calculation target of the Hough transform, a set of pixels in a narrow area having a low possibility of forming a straight line does not affect the straight line extraction. Further, it is only necessary to perform an operation on one rectangle using only four angle parameters, and the amount of operation required for one operation target can be reduced.

[0020]

As described above, according to the present invention, a set of pixels to be subjected to the Hough transform is included in a rectangle, and the rectangle is subjected to a limited number of Hough transforms. The remarkable effect that the number of operation objects and the amount of operation required for one operation object can be reduced can be obtained.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a Hough transform apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating the principle of Hough transform.

FIG. 3 is a view for explaining rectangular approximation by the inclusion processing means 1;

FIG. 4 is a view for explaining parameters required by the Hough calculation means 2.

[Explanation of symbols]

1 Inclusion processing means 2 Huff calculation means 3 Hough space memory 4 Rectangle 4a Black pixel 4b Pixel other than black pixel 5 Center point of rectangle 6 Straight line passing through the center point of the rectangle and forming an angle θ1 with the coordinate axis 7 Center point of the rectangle A straight line having an angle of θ2 with the passing coordinate axis 8 A straight line having an angle of 0 ° passing through the center point of the rectangle 9 and a straight line having an angle of 90 ° passing through the center point of the rectangular shape and the coordinate axis

Claims (2)

(57) [Claims] 1. An image processing apparatus according to claim 1, wherein a pixel constituting an image is input to a first coordinate axis orthogonal to each other in an image space.
An inclusive processing means that includes a rectangular area having sides parallel to the second coordinate axis, a Huff space memory that records the degree of overlap in the Hough space, and a first straight line obtained by extending two diagonal lines of the rectangle and
The first coordinates passing through a second straight line and a center point of the rectangle;
A third straight line parallel to the axis and the second coordinate axis;
And the fourth straight line is the first coordinate
The angle between the axis and the axis is calculated and used as the Hough transform operation angle.
Calculate the distance from the origin of the image space to each straight line
And calculate the result of the Hough transform,
The ratio of the short side is calculated by the Huff corresponding to the calculation angle and the calculation result.
Huff operator to add to address data in spatial memory
Fast Hough conversion device characterized by comprising the stage. 2. The method according to claim 1, wherein the inclusive processing means includes, for all the pixels constituting the image, starting from the pixels constituting the image not included in the rectangle, the number of pixels constituting the image and the number of pixels constituting the image. When the ratio of the number of pixels other than is greater than a specific threshold, the length of the side of the rectangle is extended in one or both coordinate axis directions of the image space so as to track and include pixels constituting an adjacent image, The length of the side when the ratio of the number of pixels constituting the image to the number of other pixels is equal to or less than the threshold value and when there are no pixels constituting the adjacent image that is not tracked and not included in the other rectangle. 2. The high-speed Hough transform apparatus according to claim 1, wherein the rectangle is generated so that the rectangle is determined, the coordinates of the vertices are output, and the rectangle having a number of pixels equal to or less than a threshold value is removed.