CN102663395B - A straight line detection method based on self-adaptation multi-scale fast discrete Beamlet transform - Google Patents

Abstract

The present invention belongs to the technical field of image processing, and particularly relates to a straight line detection method based on self-adaptation multi-scale fast discrete Beamlet transform. Technical features of the method rest with following details: an edge detection image E is obtained by carrying out edge detection on an image I using a Canny operator, gradient direction angles of edge pixel points on the edge detection image E are calculated and the edge contour is divided into various linearity regions based on a gradient direction angle of two edge points, and fast discrete Beamlet transform is carried out on a current self-adaptation partition subblock through linearity region screening to detect straight lines. The method reduces calculation amount by a certain degree because the scale of an AMFDBT algorithm initial blockette is in a self-adaptation set by the region where a target point is] in the image. At the same time, the initial Beamlet transform can easily cause straight lines to be truncated while the self-adaptive change in partition sub-block makes result of the detected straight line more complete with no truncation of straight lines.

Description

Line Detection Method Based on Adaptive Multiscale Fast Discrete Beamlet Transform

technical field

The invention belongs to image processing, in particular to a straight line detection method based on adaptive multi-scale fast discrete Beamlet transformation.

Background technique

The extraction of line features has always been a research hotspot in the field of image processing and pattern recognition. Although traditional line detection algorithms can detect straight lines quickly, they all fail because they do not provide the length information of the line segment and the starting point and end point information of the line segment. Precise positioning of line segments cannot be achieved.

Beamlet transform is one of the effective tools for multi-scale geometric analysis. It is used to restore straight lines, curves and blocky areas from noisy images. It is easy to achieve accurate positioning of line segments and multi-scale approximation of line segments. It is used in the detection of noisy line features. It has obvious advantages and potential in straight line fitting. However, the traditional Beamlet transformation requires a lot of calculations, and many improved fast algorithms waste a lot of calculations because of a lot of repetitive work. Moreover, since the block sizes of the traditional Beamlet transformation are all binary squares, it is relatively cumbersome to apply to general rectangular images. At the same time, since the algorithm is implemented on the basis of binary recursion, it is easy to cause the straight line to be truncated.

To sum up, the traditional Beamlet transformation has the disadvantages of large amount of calculation and incomplete detection results.

Contents of the invention

technical problem to be solved

In order to avoid the shortcomings of the prior art, the present invention proposes a line detection method based on adaptive multi-scale fast discrete Beamlet transform.

Technical solutions

A straight line detection method based on adaptive multi-scale fast discrete Beamlet transform, characterized in that the steps are as follows:

Step 1: Use the Canny operator to perform edge detection on the image I to obtain the edge detection image E;

当两个边缘点的梯度方向角相差大于预设阈值threshold时，在此该边缘点处断开，将边缘轮廓分割成多个线性区域并给予标记；式中，G_x和G_y分别表示边缘点的水平和垂直梯度分量；Step 2: Calculate the direction angle of the edge pixel gradient on the edge detection image E:

When the difference between the gradient direction angles of two edge points is greater than the preset threshold threshold, the edge point will be disconnected, and the edge contour will be divided into multiple linear regions and marked; where G _x and G _y represent the edge the horizontal and vertical gradient components of the point;

Step 3: Delete the line area with less than 20 edge points in the linear area to obtain the image E' filtered by the line area, and use the edge point after edge detection as the target point;

Step 4: Take the edge image E' as the initial partition sub-block, and find the smallest area where the target point exists as the adaptive partition sub-block; if the sub-block is greater than or equal to the specified threshold t', perform step 5; otherwise, perform step 6;

Step 5: Perform fast discrete Beamlet transformation on the current adaptive partition sub-block to detect straight lines. If the beamlet straight line is found, record the endpoint and slope information of the straight line, and then erase the straight line under the current sub-block; otherwise, divide the current partition block into four equal parts, and use it as the initial size of the sub-block of the next scale partition, repeat step 4 ;

Step 6: If the number of beamlet straight lines is greater than 0, according to the slope information of all straight lines, find all straight lines whose slope difference is within 1 degree as parallel lines.

Beneficial effect

A straight line detection method based on adaptive multi-scale fast discrete Beamlet transform proposed by the present invention, since the scale of the initial block of the AMFDBT algorithm is adaptively set by the area where the target point exists in the figure, in the most extreme case the target The points will spread all over the whole graph, which reduces the amount of calculation to a certain extent. At the same time, the original Beamlet transformation is likely to cause the line to be truncated, and the adaptive change of the partition sub-block makes the final detection result of the line more complete and less prone to truncation.

Description of drawings

Figure 1: Flow chart of parallel line detection based on AMFDBT

Detailed ways

Now in conjunction with embodiment, accompanying drawing, the present invention will be further described:

The present invention defines an Adaptive Tile Recursive Dyadic Partitioning (ATRDP for short) algorithm on the basis of the traditional discrete Beamlet transform binary recursive partitioning algorithm. Here S _{i, j, l} are used to represent each partition sub-block, where (i, j) represents the x-coordinate and y-coordinate of the upper left corner point of the sub-block in the original image, l represents the side length of the sub-block, and k represents the kth partition sub-blocks, and J represents the Jth adaptive tile regression partition. Then the ATRDP algorithm is defined as follows:

① ATP _J is an ATRDP.

被按其分区内目标点存在的区域自适应调整为

则②If ${\mathrm{ATP}}_J=\{S_{i_0,j_0,l_0},....S_{i_k,j_k,l_k},...,S_{i_{\mathrm{no}},j_{\mathrm{no}},l_{\mathrm{no}}}\}$ is an ATRDP, and If the adaptive partition condition is met (meaning that the beamlet b line segment can be found in the partition block and the elimination of the target point contained in the b line segment from the graph will change the distribution area of the target point in the partition block), then

It is adaptively adjusted according to the region where the target point exists in its partition as

but

${\mathrm{<span\; class="notranslate">\; ATP</span>}}_{\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; j</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; l</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; j</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; l</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; S</span>}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; adap</span>}<span\; class="notranslate">\; )</span>}_{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; j</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; l</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; j</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; l</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}<span\; class="notranslate">\; \}</span>$

Also ATRDP.

能被细分为四个相等分区子块

则分别按这四个子分区块中目标点存在的区域重新调整其尺度，得到

则③ If the adaptive partition condition is not met, and

Can be subdivided into four equal partition subblocks

Then readjust the scale according to the area where the target point exists in the four sub-blocks, and get

but

是

的父块。 ${\mathrm{ATP}}_{J+1}=\begin{array}{c}\{S_{i_0,j_0,l_0},...,S_{i_k,j_k,l_k},...,S{\left(\mathrm{adap}\right)}_{i_k,j_k,l_k/2},S{\left(\mathrm{adap}\right)}_{i_k+l_k/2,j_k,l_k/2},{S\left(\mathrm{adap}\right)}_{i_k,j_k+l_k/2,l_k/2},\\ S{\left(\mathrm{adap}\right)}_{i_k+l_k/2,j_k+l_k/2,l_k/2...,S_{i_{\mathrm{no}},j_{\mathrm{no}},l_{\mathrm{no}}}\}}\end{array}$ is also ATRDP, where

yes

parent block.

Combining the ATRDP partition method with the Fast Discrete Beamlet Transform (FDBT for short), the adaptive multi-scale Fast Discrete Beamlet Transform is defined as:

为快速离散Beamlet变换，定义如下：in

For the fast discrete Beamlet transform, it is defined as follows:

$\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; t</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; t</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

(其中||b||是beamlet b上的元素个数)超过指定阈值t^*的beamlet b，在上一层大尺度下子块已经存在的beamlet b在下一级分解的精细尺度子块中将不再考虑，大尺度下的beamlet b的分解可以表示为b＝U_jb_j，所以AMFDBT的截集为：In the formula (1), beamlet b is a line segment in the partition sub-block S, and S is the ATRDP partition sub-block. The cut set of AMFDBT includes all energy functions

(Where ||b|| is the number of elements on beamlet b) Beamlet b exceeding the specified threshold t ^* , beamlet b that already exists in sub-blocks at the upper level of large-scale sub-blocks will not be decomposed in the fine-scale sub-blocks of the next level Considering again, the decomposition of beamlet b on a large scale can be expressed as b=U _j b _j , so the cut set of AMFDBT is:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; t</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; ATP</span>}}_{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&NotElement;</span>{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

${\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; t</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

where P is the parent block of S.

(2) Parallel line detection based on AMFDBT

The invention applies the AMFDBT to the parallel line detection of the image.

Before detecting parallel lines, a series of preprocessing is first performed on the image, and the specific steps are as follows:

① Use the Canny operator to detect the edge of the image I to obtain the edge detection image E.

② Linear area screening, eliminating curves or shorter straight lines. Because in image I, besides the target, there are often a large number of complex backgrounds, after edge detection, they mainly show some irregular or curved edge features, and these edge features will have a greater impact on the straight line detection process. Therefore, before the straight line detection, the edge profile of the edge image E is screened first, leaving only the linear area with more edge points in the same direction, and removing some curved irregular or too short edge features.

The present invention mainly utilizes the gradient phase information of edge pixels to realize the screening of linear regions. First calculate the orientation angle of the edge pixel gradient:

where G _x and G _y represent the horizontal and vertical gradient components of edge points, respectively. When the difference between the gradient orientation angles of two edge points is greater than the preset threshold threshold, it means that the direction of the edge contour changes greatly, and the edge point is disconnected. In this way, the edge contour can be divided into many linear areas, these linear areas are marked, the number of edge points in the linear area is counted, and the line areas with fewer edge points are deleted to further reduce detection interference. We denote the image filtered by the line area as E'.

Using AMFDBT to detect parallel lines on the basis of edge image E′, the specific algorithm is described as follows:

① Traverse the operation subgraph (the initial operation subgraph is the edge image E′), and find the area where the target point (ie, the edge point of the image after edge detection) exists in the graph. If the area size is greater than the specified threshold t′ (t′ can be adjusted according to the image size, generally image width/100), go to the next step; otherwise go to step ④.

②Using the target point area found in step ① as the partition sub-block, perform fast discrete Beamlet transformation on the partition block, find out all the beamlet straight lines in the current partition block, and calculate their slope according to the endpoint information of the straight line.

③If the beamlet straight line is found in step ②, extract the straight line information and erase the straight line; otherwise, divide the current partition block into four equal parts and use it as the initial size of the sub-block of the next scale partition. Execute step ①.

④ If the beamlet straight line is detected, all parallel lines are found according to the slope information of the straight line.

Referring to the flow chart (1) in the embodiment of the present invention, the parallel line detection steps based on adaptive multi-scale Beamlet transformation are as follows:

The present invention applies AMFDBT to carry out parallel line detection to image, and its specific steps are as follows:

(1) Image preprocessing

① Use the Canny operator to detect the edge of the image I to obtain the edge detection image E.

② Linear area screening, eliminating curves or shorter straight lines. Because in image I, besides the target, there are often a large number of complex backgrounds, after edge detection, they mainly show some irregular or curved edge features, and these edge features will have a greater impact on the straight line detection process. Therefore, before the straight line detection, the edge profile of the edge image E is screened first, leaving only the linear area with more edge points in the same direction, and removing some curved irregular or too short edge features.

The present invention mainly utilizes the gradient phase information of edge pixels to realize the screening of linear regions. First calculate the orientation angle of the edge pixel gradient:

where G _x and G _y represent the horizontal and vertical gradient components of edge points, respectively. When the difference between the gradient direction angles of two edge points is larger than the preset threshold threshold (threshold=20), it means that the direction of the edge contour changes greatly, and the edge point is disconnected. In this way, the edge contour can be divided into many linear areas, these linear areas are marked, the number of edge points in the linear area is counted, and the line areas with fewer edge points are deleted to further reduce detection interference. We denote the image filtered by the line area as E'.

(2) Parallel line detection based on AMFDBT

① Traverse the operation subgraph (the initial operation subgraph is the edge image E′), and find the area where the target point (ie, the edge point of the image after edge detection) exists in the graph. If the area size is greater than the specified threshold t′ (t′ can be adjusted according to the image size, generally image width/100), go to the next step; otherwise go to step ④.

②Using the target point area found in step ① as the partition sub-block, perform fast discrete Beamlet transformation on the partition block, find out all the beamlet straight lines in the current partition block, and calculate their slope according to the endpoint information of the straight line.

③ If the beamlet straight line is found in step ②, extract the straight line information (end point and slope) and erase the straight line; otherwise, divide the current partition block into four equal parts and use it as the initial size of the sub-block of the next scale partition. Execute step ①.

④ If the beamlet straight line is detected, all parallel lines are found according to the slope information of the straight line.

Claims (1)

1. line detection method based on self-adapting multi-dimension fast discrete Beamlet conversion is characterized in that step is as follows:

Step 1: utilize the Canny operator to carry out rim detection to image I, obtain edge-detected image E;

Step 2: the deflection of edge calculation detected image E up contour point gradient:

When the gradient direction angle of two marginal points differs by more than predetermined threshold value threshold, expression edge contour direction transformation is larger, judge that these two marginal points do not belong to the same range of linearity, are divided into several ranges of linearity and separate marking with this with all marginal points; In formula, G _xAnd G _yThe horizontal and vertical gradient component that represents respectively marginal point;

Step 3: all are contained the edge count and be less than the deletion of 20 the range of linearity, obtain through the image E ' after the screening of the range of linearity, and with edge image E ' as subregion sub-block at first, with all marginal points in E ' as impact point;

Step 4: seek Minimum Area that impact point exists as the adaptive partition sub-block under current subregion sub-block, if the size of this adaptive partition sub-block during more than or equal to predetermined threshold value t', execution in step 5; Otherwise execution in step 6;

Step 5: current adaptive partition sub-block is carried out fast discrete Beamlet conversion, detection of straight lines; If find the beamlet straight line, record end points and the slope information of straight line, then wipe this straight line under current subregion sub-block; Otherwise, with the current subregion sub-block quartern, and as the initial size of next yardstick subregion sub-block, turn back to step 4;

Step 6: if the quantity of beamlet straight line greater than 0, according to the slope information of all straight lines, find straight slope poor 1 the degree with all interior straight lines as parallel lines.

Images