JPS6284391A - Long and thin rectangular extraction method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Summary] This is a method for extracting elongated rectangles in an image processing device, in which after extracting line elements on the longer side of each rectangle, a parallel line internal region is extracted perpendicular to the parallel lines. This method detects the position of the shorter side based on the projected image.This allows accurate extraction even when a plurality of elongated rectangles exist.

[Industrial Field of Application] The present invention relates to a rectangle extraction method in an image processing device, and particularly to a method that can extract all of a plurality of elongated rectangles when they exist in an image.

[Prior Art] A conventional method for extracting a long and thin rectangle in an image involves extracting candidate line elements that make up each side of the rectangle, and then detecting the outline of the rectangle by fitting straight lines to each side. There is a method called.

[Problems to be Solved by the Invention] In the conventional rectangle extraction method described above, when the rectangle is long and thin, there are few line elements corresponding to the short sides of the rectangle, so it is difficult to extract the sides. Also on the side, 2
Since the line elements of the parallel sides of the book are close to each other, there is a problem in that it is not easy to fit straight lines for each side.

In particular, when a plurality of rectangles exist in an image, the above problem becomes more serious.

The present invention aims to provide a novel elongated rectangle extraction method that solves these problems.

[Means for Solving the Problems] FIG. 1 shows a block diagram of the principle of the elongated rectangle extraction method of the present invention.

FIG. 1(a) is a block diagram of the entire elongated rectangle extraction method of the present invention.

In FIG. 1(a), reference numeral 1 denotes a line element extraction unit that extracts the longer line element of each rectangle in the input image.

2 is for the line element extracted by the line element extraction unit 1,
This is a parallel line pair detection unit that matches a pair of parallel lines with a small width to each rectangle.

Reference numeral 3 denotes a projection unit that projects the internal region of the parallel lines in a direction perpendicular to the parallel lines for each pair of parallel lines detected by the parallel line pair detection unit 2.

Reference numeral 4 denotes a rectangular end point detection section that extracts points where the degree of change in projection value is large for each projection image obtained by the projection section 3, and a rectangle is extracted by detecting the rectangular end points.

The line element extraction unit 1 can be realized by the configuration shown in the 17th mountain).

To input an image, first, the differential value of each pixel in the X direction and in the X direction is obtained in the differential detection section 11, and the magnitude and direction of the differential are detected.

Next, the line element candidate detection unit 12 detects pixels whose differential direction is within a preset range and whose differential magnitude is greater than or equal to a preset threshold as line element candidates.

Next, the non-maximum suppression unit 13 deletes, among the detected line element candidates, those whose differential values are not maximal in the vicinity of a predetermined size on a straight line along the direction of differentiation.

As a result, only the pixel with the maximum density gradient is left and output as a line element.

The parallel line pair detection section 2 can be realized by a configuration as shown in FIG. 1(C).

A Hough transformation unit 21 performs Hough transformation on each pixel constituting the line element extracted by the line element extraction unit 1 to create a histogram of straight line parameters.

That is, for each pixel, the parameters of all possible straight lines passing through that pixel are determined, and a histogram regarding the parameters is created.

Next, in the candidate straight line parameter detection unit 22, H
Regarding the histogram of straight line parameters created by the ough conversion unit 21, a parameter of a straight line having the maximum frequency in the vicinity of a certain size and a predetermined threshold value or more is determined.

Next, the parallel line pair parameter detection section 23 detects the parameters of a straight line that is parallel to each of the candidate straight lines detected by the candidate straight line parameter detection section 22 and that has the largest frequency in the histogram in the vicinity.

This creates the parameters of the pair of parallel lines that form the longer sides of the elongated rectangle.

[Operation] With the above configuration, after extracting the line elements on the longer side of each rectangle in the image, a pair of parallel lines with a small width is fitted to each rectangle, and the interior of the parallel line is calculated for each pair of parallel lines. The area is projected in a direction perpendicular to parallel lines, and the position of the shorter side of the rectangle is detected based on the projected image.

Further, in detecting parallel lines, first the most probable straight line is detected for each rectangle, and then another straight line parallel to that straight line is detected in the vicinity of that straight line.

This makes it possible to correctly extract all of the plurality of elongated rectangles in the image.

[Example] The present invention will be described in more detail below with reference to Examples shown in FIGS. 2 and 3.

FIG. 2 is a block diagram of an embodiment of the invention.

The operation of this embodiment will be explained according to FIG.

(1) Image memory 6 holds input image data.

The input image data is 512x512x8 bit digital image data.

(2) The x-direction differentiator 111 and the y-direction differentiator 112 scan the image with a 3×3 spatial filter shown in FIGS. 3(a) and 3(b), respectively. Through this scanning, differential values as shown in FIGS. 3(C) and 3(d) are obtained, respectively.

(3) The differential magnitude detection unit 113 is the X-direction differentiator 11
1 and the square root of the sum of squares of the X-direction differential value and the X-direction differential value obtained by the y-direction differentiator 112. The differential direction detection unit 114 calculates the direction of the vector created by the differential value in the X direction and the differential value in the X direction. The direction setting unit 115 sets an approximate value of the direction of the longer side of the elongated rectangle.

(4) The line element candidate detection unit 12 uses the differential direction detection unit 114
The direction of the detected differential is almost perpendicular to the direction set by the direction setting section 115, and the magnitude of the differential calculated by the differential magnitude detection section 113 is greater than or equal to a predetermined threshold. To detect.

(5) Among the line element candidates, the non-maximum suppression unit 13 deletes a pixel whose differential value is not the maximum in the vicinity of a certain size centered on the pixel on a straight line along the direction of differentiation.

(6) For each pixel of the remaining line element candidates in the non-maximum suppression unit I3, the Hough conversion unit 21 converts all possible straight line parameters passing through that pixel: A, B (Ax
+By=1) and create a histogram regarding that parameter.

(7) The candidate straight line parameter detection unit 22 calculates
In the histogram of straight line parameters created by the h-conversion unit 21, the frequency of the histogram is greater than or equal to a preset threshold, and the frequency is maximum in the vicinity of a predetermined size centered on the parameter. Detect the parameters of the straight line. Through this process, the most probable straight line is detected one by one for each rectangle.

(8) Parallel line pair parameter detection unit 23 detects a line that is parallel to each candidate straight line detected by candidate straight line parameter detection unit 22 and whose histogram frequency is maximum near a predetermined size. Detect the parameters of the straight line.

(9) For each parallel line pair detected by the parallel line pair parameter detection unit 23, the projection unit 3 projects the internal region thereof in a direction perpendicular to the parallel line based on the data from the image memory 6.

(10) The rectangle end point detection unit 4 detects the position of the shorter side of the rectangle by extracting a point where the degree of change in projection value is large for each projection image projected by the projection unit 3.

(11) The rectangular parameter detection unit 5 detects the parameters of the parallel line pair detected by the parallel line pair parameter detection unit 23,
The parameters of each rectangle are detected from the position of the shorter side of the rectangle detected by the rectangle end point detection unit 4.

[Effects of the Invention] As described above, according to the present invention, even when a plurality of elongated rectangles exist, these rectangles can be correctly extracted, and the effect in image processing is extremely large.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, and FIG. 3 is an explanatory diagram of differential processing. In the drawings, 1 is a line element extraction section, 2 is a parallel line pair detection section, 3 is a projection section, 4 is a rectangular end point detection section, 5 is a rectangular parameter detection section, 6 is an image memory, 11 is a differential detection section, 12 13 is a line element candidate detection unit, 13 is a non-maximum suppression unit, 21 is a Hough transformation unit, 22 is a candidate straight line parameter detection unit, and 23 is a parallel line pair parameter detection unit, respectively. Extra human power;〕kol〕2にsquare〉jchibu〉搗BM ”)'f 1”bu'o-,,7 1,000 3 Cb) Ekiji (2) Figure

Claims (3)

[Claims] (1) A line element extraction unit (1) that extracts line elements on the longer side of each rectangle in the input image; a parallel line pair detection unit (2) that fits a pair of parallel lines with a small width for each rectangle; a projection unit (3) that projects an internal region of the parallel line in a direction perpendicular to the parallel line for each pair of parallel lines; An elongated rectangle extraction method comprising: a rectangle end point detection section (4) for extracting points with a large degree of change in projection value for each projection image obtained by the projection section (3). (2) The parallel line pair detection unit (2) includes a Hough transformation unit (21) that performs Hough transformation on pixels constituting a line element and creates a histogram of straight line parameters; a candidate straight line parameter detection unit (22) for determining the parameters of a straight line that has a maximum frequency in the vicinity of a predetermined size and whose frequency is greater than or equal to a predetermined threshold value in the histogram of straight line parameters created by , parallel line pairs for detecting parameters of straight lines that are parallel to each candidate straight line detected by the candidate straight line parameter detection unit (22) and whose frequency in the histogram is maximum near a predetermined size. The elongated rectangle extraction method according to claim 1, characterized in that the elongated rectangle extraction method is configured to include a parameter detection section (23). (3) The line element detecting unit (1) includes a differential detecting unit (11) that detects the magnitude and direction of the differential in each pixel, and the direction of the differential is within a preset range, and A line element candidate detection unit (12) detects pixels for which the magnitude of the differential is equal to or greater than a predetermined threshold; A non-maximum control unit (13
) The elongated rectangle extraction method according to claim 1, characterized in that it is configured to include the following.