CN102663395A - 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

The line detection method converted based on self-adapting multi-dimension fast discrete Beamlet

Technical field

The invention belongs to image procossing, more particularly to a kind of line detection method converted based on self-adapting multi-dimension fast discrete Beamlet.

Background technology

The extraction of line feature is always a study hotspot in image procossing and area of pattern recognition, although traditional line detection algorithm can detect straight line quickly, all because being accurately positioned for line segment can not be realized without the length information for providing line segment and the beginning and end information of line segment.

Beamlet conversion is one of effective tool of multi-scale geometric analysis, for recovering straight line from noisy image, curve and boxed area, it easily realizes being accurately positioned and to the multiple dimensioned approximate of line segment, obvious advantage and potentiality being suffered from the detection of noisy linear feature and fitting a straight line for line segment.But the amount of calculation of tradition Beamlet conversion is very big, many improved fast algorithms all waste many amounts of calculation because of to carry out substantial amounts of repeated work.And enter square because traditional Beamlet subregion block sizes converted are two, to then relatively cumbersome applied to general rectangular image.Simultaneously as algorithm two enter it is recursive on the basis of realize, easily cause the situation that straight line is truncated.

In summary, there is the deficiencies such as computationally intensive, testing result is imperfect in traditional Beamlet conversion.

The content of the invention

The technical problem to be solved

In order to avoid in place of the deficiencies in the prior art, the present invention proposes a kind of line detection method converted based on self-adapting multi-dimension fast discrete Beamlet.

Technical scheme

A kind of line detection method converted based on self-adapting multi-dimension fast discrete Beamlet, it is characterised in that step is as follows：

Step 1：Rim detection is carried out to image I using Canny operators, edge-detected image E is obtained；

Step 2：Calculate the deflection of edge-detected image E top edge pixel gradients：When the gradient direction angle of two marginal points differs by more than predetermined threshold value threshold, disconnected herein at the marginal point, edge contour is divided into multiple ranges of linearity and given mark；In formula, G_xAnd G_yThe horizontal and vertical gradient component of marginal point is represented respectively；

Step 3：The line area that range of linearity inward flange is counted less than 20 is deleted, and obtains the image E ' after line area is screened, and regard the marginal point after rim detection as target point；

Step 4：With edge image E ' for initial subregion sub-block, the Minimum Area for finding target point presence is used as adaptive partition sub-block；If the sub-block performs step 5 when being more than or equal to defined threshold t '；Otherwise step 6 is performed；

Step 5：Fast discrete Beamlet conversion, detection of straight lines are carried out to current adaptive partition sub-block.If finding beamlet straight lines, the end points and slope information of straight line are recorded, the straight line is then wiped under current sub-block；Otherwise, by the current bay block quartering, and as the initial size of next yardstick subregion sub-block, step 4 is repeated；

Step 6：If the quantity of beamlet straight lines is more than 0, according to the slope information of all straight lines, all straight lines of the straight slope difference within 1 degree are found as parallel lines.

Beneficial effect

A kind of line detection method converted based on self-adapting multi-dimension fast discrete Beamlet proposed by the present invention, because the yardstick of AMFDBT algorithm primary partition blocks is that the region adaptivity existed by target point in figure is set, target point can just spread all over full figure in the most extreme case, and this make it that amount of calculation obtains certain reduction.Meanwhile, initial Beamlet conversion easily causes the situation that straight line is truncated, and the change of subregion sub-block self adaptation make it that the result of final detection of straight lines is more complete, is not likely to produce and blocks phenomenon.

Brief description of the drawings

Fig. 1：Parallel lines overhaul flow chart based on AMFDBT

Embodiment

In conjunction with embodiment, accompanying drawing, the invention will be further described：

The present invention defines self adaptation tile on the basis that recursive partitioning algorithm is entered in traditional discrete Beamlet conversion two and returns subregion (Adaptive Tile Recursive Dyadic Partitioning, abbreviation ATRDP) algorithm.Here S is used_{I, j, l}Each subregion sub-block is represented, wherein (i, j) represents the x coordinate and y-coordinate of the sub-block upper left angle point in original image, l represents the length of side of sub-block, and k represents k-th of subregion sub-block, and J represents to carry out self adaptation tile recurrence subregion the J times.Then ATRDP algorithms are defined as follows：

①ATP_JIt is an ATRDP.

If 2. ${\mathrm{ATP}}_J=\{S_{i_0,j_0,l_0},....S_{i_k,j_k,l_k},...,S_{i_n,j_n,l_n}\}$ It is an ATRDP, and

If meeting adaptive partition condition (refer to find beamlet b line segments in the blockette and eliminate the target point contained by b line segments from figure and can change the distributed areas of target point in blockette),

It is adjusted to by the region adaptivity existed by its intrapartition destination punctuate

Then

${\mathrm{ATP}}_{J+1}=\{S_{i_0,j_0,l_0},...,S_{i_k,j_k,l_k},S{\left(\mathrm{adap}\right)}_{i_k,j_k,l_k},...,S_{i_n,j_n,l_n}\}$

It is also ATRDP.

③If adaptive partition condition is unsatisfactory for, and

Four equal divisions sub-blocks can be subdivided into

Its yardstick is readjusted in the region then existed respectively by target point in this four sub- blockettes, is obtained

Then

${\mathrm{ATP}}_{J+1}=\begin{array}{c}\{S_{i_0,j_0,l_0},...,S_{i_k,j_k,l_k},...,S{\left(\mathrm{adap}\right)}_{i_k,j_k,l_k/2},S{\left(\mathrm{adap}\right)}_{i_k+l_k/2,j_k,l_k/2},{S\left(\mathrm{adap}\right)}_{i_k,j_k+l_k/2,l_k/2},\\ S{\left(\mathrm{adap}\right)}_{i_k+l_k/2,j_k+l_k/2,l_k/2...,S_{i_n,j_n,l_n}\}}\end{array}$ It is also ATRDP, wherein

It is

Father's block.

ATRDP partition methods and fast discrete Beamlet conversion (Fast Discrete Beamlet Transform, abbreviation FDBT) are combined, self-adapting multi-dimension fast discrete Beamlet is defined and is transformed to：

WhereinConvert, be defined as follows for fast discrete Beamlet：

$G\left(f(x,y)\right)=\left\{\begin{array}{cc}1& f(x,y)\&GreaterEqual;t\\ 0& f(x,y)<t\end{array}\right.---\left(3\right)$

Beamlet b are a line segment in subregion sub-block S in formula (1), and S is ATRDP subregion sub-blocks.AMFDBT cut set includes all energy functions

(wherein | | b | | it is the element number on beamlet b) exceed specified threshold t^*Beamlet b, the beamlet b that sub-block has been present under last layer large scale will not be considered any further in the fine dimension sub-block that next stage is decomposed, and the decomposition of the beamlet b under large scale can be expressed as b=U_jb_j, so AMFDBT cut set is：

$B_{t^{*}}=\left\{b\right|E\left(b\right)>t^{*},|b\&Element;S,S\&Element;{\mathrm{ATP}}_J,b\&NotElement;{\tilde{D}}_{t^{*}}\}---\left(4\right)$

${\tilde{D}}_{t^{*}}=\left\{b_j\right|b=U_j{b}_j,E\left(b\right)>t^{*},b\&Element;P\}---\left(5\right)$

Wherein P is S father's block.

(2) the parallel lines detection based on AMFDBT

AMFDBT is applied to carry out parallel lines detection to image by the present invention.

First have to carry out series of preprocessing to image before parallel detection, it is comprised the following steps that：

1. rim detection is carried out to image I using Canny operators, obtains edge-detected image E.

2. the range of linearity is screened, and rejects curve or compared with short straight line.Due in image I, in addition to target there are after substantial amounts of complex background, rim detection them toward contact mainly shows a little irregular or bending edge feature, and these edge features can cause large effect in straight-line detection process.Therefore before straight-line detection, edge contour screening first is carried out to edge image E, the more range of linearity of marginal point in the same direction is left behind, removed some and bend irregular or too short edge feature.

The present invention mainly uses the Gradient Phase information of edge pixel point, to realize the screening of the range of linearity.The deflection of edge pixel point gradient is calculated first：

In formula, G_xAnd G_yThe horizontal and vertical gradient component of marginal point is represented respectively.When the gradient direction angle of two marginal points differs by more than predetermined threshold value threshold, represent that edge contour direction transformation is larger, disconnected at this marginal point.Edge contour can be so divided into many ranges of linearity, these ranges of linearity are marked, statistically linear region inner margin points delete the less line area of marginal point, detection interference is reduced with further.Image after line area is screened is designated as E ' by us.

AMFDBT parallel detections are utilized on the basis of edge image E ', specific algorithm is described as follows：

1. traversing operation subgraph (initial operation subgraph is edge image E '), finds the region that target point in figure (i.e. marginal point of the image after rim detection) is present.If the area size is more than defined threshold t ', (t ' can be adjusted according to image size, generally picture traverse/100), perform next step；Otherwise step is performed 4..

2. the target point region 1. found using step is carried out fast discrete Beamlet conversion to the blockette, finds out all beamlet straight lines in the block of current bay, its slope is calculated according to straight line terminal point information as subregion sub-block.

If 3. step 2. in find beamlet straight lines, extract straight line information and simultaneously wipe the straight line；Otherwise, by the current bay block quartering, and it is used as the initial size of next yardstick subregion sub-block.Perform step 1..

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

Embodiment of the present invention reference flow sheet (1), the parallel lines detecting step converted based on self-adapting multi-dimension Beamlet is as follows：

AMFDBT is applied to carry out parallel lines detection to image by the present invention, and it is comprised the following steps that：

(1) image preprocessing

1. rim detection is carried out to image I using Canny operators, obtains edge-detected image E.

2. the range of linearity is screened, and rejects curve or compared with short straight line.Due in image I, in addition to target there are after substantial amounts of complex background, rim detection them toward contact mainly shows a little irregular or bending edge feature, and these edge features can cause large effect in straight-line detection process.Therefore before straight-line detection, edge contour screening first is carried out to edge image E, the more range of linearity of marginal point in the same direction is left behind, removed some and bend irregular or too short edge feature.

The present invention mainly uses the Gradient Phase information of edge pixel point, to realize the screening of the range of linearity.The deflection of edge pixel point gradient is calculated first：

In formula, G_xAnd G_yThe horizontal and vertical gradient component of marginal point is represented respectively.When the gradient direction angle of two marginal points differs by more than predetermined threshold value threshold (threshold=20 herein), represent that edge contour direction transformation is larger, disconnected at this marginal point.Edge contour can be so divided into many ranges of linearity, these ranges of linearity are marked, statistically linear region inner margin points delete the less line area of marginal point, detection interference is reduced with further.Image after line area is screened is designated as E ' by us.

(2) the parallel lines detection based on AMFDBT

1. traversing operation subgraph (initial operation subgraph is edge image E '), finds the region that target point in figure (i.e. marginal point of the image after rim detection) is present.If the area size is more than defined threshold t ', (t ' can be adjusted according to image size, generally picture traverse/100), perform next step；Otherwise step is performed 4..

2. the target point region 1. found using step is carried out fast discrete Beamlet conversion to the blockette, finds out all beamlet straight lines in the block of current bay, its slope is calculated according to straight line terminal point information as subregion sub-block.

If 3. step 2. in find beamlet straight lines, extract straight line information (end points and slope) and simultaneously wipe the straight line；Otherwise, by the current bay block quartering, and it is used as the initial size of next yardstick subregion sub-block.Perform step 1..

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

Claims (1)

1. a kind of line detection method converted based on self-adapting multi-dimension fast discrete Beamlet, it is characterised in that step is as follows：

Step 1：Rim detection is carried out to image I using Canny operators, edge-detected image E is obtained；

Step 2：Calculate the deflection of edge-detected image E top edge pixel gradients：

, when the gradient direction angle of two marginal points differs by more than predetermined threshold value threshold, disconnected herein at the marginal point, edge contour is divided into multiple ranges of linearity and given mark；In formula, G_xAnd G_yThe horizontal and vertical gradient component of marginal point is represented respectively；

Step 3：The line area that range of linearity inward flange is counted less than 20 is deleted, and obtains the image E ' after line area is screened, and regard the marginal point after rim detection as target point；

Step 4：With edge image E ' for initial subregion sub-block, the Minimum Area for finding target point presence is used as adaptive partition sub-block；If the sub-block performs step 5 when being more than or equal to defined threshold t '；Otherwise step 6 is performed；

Step 5：Fast discrete Beamlet conversion, detection of straight lines are carried out to current adaptive partition sub-block.If finding beamlet straight lines, the end points and slope information of straight line are recorded, the straight line is then wiped under current sub-block；Otherwise, by the current bay block quartering, and as the initial size of next yardstick subregion sub-block, step 4 is repeated；

Step 6：If the quantity of beamlet straight lines is more than 0, according to the slope information of all straight lines, all straight lines of the straight slope difference within 1 degree are found as parallel lines.