CN104156956A - Multi-angle edge detection method based on Gauss wavelet one-dimensional peak value identification - Google Patents

Abstract

The invention relates to a multi-angle edge detection method based on Gauss wavelet one-dimensional peak value identification. The multi-angle edge detection method comprises the following steps: histogram analysis is performed on a to-be-processed image to obtain the gray value of a target and a background of the target, and the obtained gray value is adopted to perform assignment on 18 given template images; multiple groups of edge detection parameters are preset, and the preset multiple groups of edge detection parameters are adopted to perform edge detection on the 18 given template images subjected to assignment, the edge detection results having the highest degree of approximation with the template images are found out, the edge detection parameters used by the edge detection results serve as the optimal parameters, the obtained optimal parameters are adopted to carry out the following items on the to-be-processed image: image segmentation, constructing multiple one-dimensional vectors, carrying out convolution operation and calculating the absolute value, determining the local maximum value, assigning gray value to the local maximum value, substituting pixels on the same position of an original image and performing binarization treatment on the edges of the images subjected to multiple superposition, so that the image subjected to edge detection can be obtained.

Description

A kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value

Technical field

The present invention relates to a kind of image processing method, particularly about a kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value.

Background technology

Picture strip, to vivid thinking world of the mankind, is the important channel in the human knowledge world.We are referred to as edge to be present in sudden change in image and discontinuous jiggly structure.Edge is often carrying abundant image information.These marginal points have formed contour of object, and these profiles are often the interested places of researcher, the feature of its concentrated reflection goal in research, to follow-up image cut apart, the identification of images match, target, computer vision has very important place mat effect, becomes so how structural unstable edge in real image is refined into constitutionally stable edge the direction that people further investigate always for many years.In the research of decades, people constantly introduce mathematical method and image border are extracted and explain.Wait until LoG operator and Canny operator by initial the Prewitt operator based on gradient principle, Sobel operator, then to wavelet transformation, embodied the degree of depth and the difficulty of rim detection problem.

Multi-angle edge detection method generally comprises template Operator Method and the two-dimensional wavelet transformation modulus maximum edge detection method based on gradient principle.Template Operator Method based on gradient principle adopts the gradient template of N*N to carry out convolution to two dimensional image.Because template is generally square, and its size is 5 pixel * 5 pixels to the maximum, so the gradient direction that template can generate mostly is 16 most, 0 °, 30 °, 45 °, 60 °, 90 °, 120 °, 135 °, 150 °, 180 °, 210 °, 225 °, 240 °, 270 °, 300 °, 315 ° and 330 ° of directions.Classical two-dimensional wavelet transformation modulus maximum edge detection method can only be sorted out and carry out non-maximum value inhibition according to argument again after asking gradient with y direction in the x-direction.Therefore the mode of utilizing existing multi-angle edge detection method to detect basic dependence image rotating and rotational coordinates to image border realizes.But all can carry out interpolation arithmetic to image when image rotating and rotational coordinates, cause the change of gradation of image information, thereby when the edge of image identification after image rotating and rotational coordinates, cannot ensure the accuracy of image border, and carrying out also needing the angle according to rotation that edge image is rotated back to original position after edge identification, cause so again the change of edge image information.In addition, image rotating and rotational coordinates can cause change and the generation image boundary problem of picture size, and these problems all can increase image intractability.

The selection of edge detection threshold is to reach according to statistical decision point that the efficient frontier of failing to detect and mistake cause noise to be classified as of marginal point compromise.The method that existing threshold value is selected comprises artificial selection method and automatic threshold method.Artificial selection method is the simplest, can pass through eye-observation result, and repeatedly image border is determined in interactive operation.But in image, the insensitive details of human eye perceives is likely left in the basket.Automatically selecting threshold method is automatically to judge according to certain decision-making technique threshold value.It is applicable to major applications environment, has difficulties but find suitable decision policy and evaluate its validity.About the also ununified standard of choosing at present of edge detection algorithm parameter, be to be applied as standard and to determine algorithm parameter whether to meet substantially.

Summary of the invention

For the problems referred to above, the object of this invention is to provide a kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value that can solve single pixel multi-angle rim detection problem under the prerequisite that does not change image information.

For achieving the above object, the present invention takes following technical scheme: a kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value, it comprises the following steps: 1) pending image is carried out to histogram analysis, obtain the gray-scale value of target and background thereof, adopt the gray-scale value obtaining to carry out assignment to 18 kinds of given template images; 2) default some groups of rim detection parameters, adopt some groups of default rim detection parameters respectively 18 kinds of template images after assignment to be carried out to rim detection, and find out the edge detection results the highest with the template image degree of approximation, to obtain this result rim detection parameter used as optimized parameter, it specifically comprises the following steps: (1) utilize m capable × image subblock of mtan θ row cuts apart template image, by template image be divided into several m capable × image subblock of mtan θ row, wherein, θ is rim detection angle, θ ∈ [45 °, 90 °); (2) m capable × build some one-dimensional vector along its diagonal line in the image subblock of mtan θ row; (3) by the some one-dimensional vector that build respectively with the first order derivative f of Gaussian function _σ(t) make convolution algorithm, and convolution algorithm result is taken absolute value; (4) determine the local maximum of convolution results absolute value; (5) local maximum that obtains is composed to gray-scale value, and according to pixel subscript by the pixel in same position in the edge image pixel replacement original image generating; (6) the some edge images that different edges detection angles direction obtained carry out gray scale stack, according to the grey level histogram of image after repeatedly superposeing, using gray-scale value the highest gray scale number percent as binary-state threshold, according to this binary-state threshold, the edge of image after repeatedly superposeing is carried out to binary conversion treatment; (7) adopt Pratt rim detection quality factor method that image border in the image border of generation and template image is compared, obtain the one group edge detection results the highest with the template image edge degree of approximation, and will obtain this edge detection results rim detection parameter used as optimized parameter; 3) adopt step 2) optimized parameter that obtains, repeating step 2) in step (1)～step

(6), pending image is carried out to rim detection.

Described step 2) in, default rim detection parameter comprises the first order derivative f of Gaussian function _σ(t) parameter σ, one dimension peak value maximum value threshold alpha and binary-state threshold.

Described step 2) in, to m capable × image subblock of mtan θ row builds one-dimensional vector, it comprises: 1. according to the span of rim detection angle θ: θ ∈ [45 °, 90 °), provide m capable × three kinds of forms of the image subblock of mtan θ row and Q _{90 °}, and Q _{90 °}be expressed as with pixel a respectively:

θ ₁for the form of image subblock corresponding rim detection angle,

θ ₂for the form of image subblock corresponding rim detection angle,

2. for m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ₁, X ₂x _mand Y ₁y _m-1,

$\left\{\begin{array}{c}{X}_1=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2k}& ...& a_{2(2k-1)},& ...& ...& a_{m[(m-1)(k-1)+1]}& ...& a_{m[m(k-1)+1]}\end{array}\right]\\ X_2=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3k}& ...& a_{3(2k-1)},& ...& ...& ,a_{m[(m-2)(k-1)+1]}& ...& a_{m[(m-1)(k-1)+1]}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}=\left[\begin{array}{ccccccccccc}{a}_{1k}& ...& a_{1(2k-1)},& a_{2(2k-1)}& ...& a_{2(3k-2)},& ...& ...& ,a_{(m-1)[(m-1)(k-1)+1]}& ...& a_{(m-1)[m(k-1)+1]}\end{array}\right]\\ Y_1=\left[\begin{array}{ccc}{a}_{1[(m-1)(k-1)+1]}& ...& a_{1[m(k-1)+1]}\end{array}\right]\end{array}\right.,$

For m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ' ₁, X' ₂x' _mand Y ' ₁y ' _m-1,

$\left\{\begin{array}{c}{X}_1^{\′}=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2(k+1)}& ...& a_{2\left(2k\right)},& ...& ...& a_{m\left[(m-1)(k-1\right]}& ...& a_{m\left(\mathrm{mk}\right)}\end{array}\right]\\ X_2^{\′}=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3(k+1)}& ...& a_{3\left(2k\right)},& ...& ...& ,a_{m[(m-2)k+1]}& ...& a_{m\left[(m-1)k\right]}\end{array}\right]\\ X_m^{\′}=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}^{\′}=\left[\begin{array}{ccccccccccc}{a}_{1(k+1)}& ...& a_{1\left(2k\right)},& a_{2(2k+1)}& ...& a_{2\left(3k\right)},& ...& ...& a_{(m-1)[(m-1)k+1]}& ...& a_{(m-1)\left(\mathrm{mk}\right)}\end{array}\right]\\ \·\\ \·\\ \·\\ Y_1^{\′}=\left[\begin{array}{ccc}{a}_{1[(m-1)k+1]}& ...& a_{1\left(\mathrm{mk}\right)}\end{array}\right]\end{array}\right.;$

For m capable × the form Q of the image subblock of mtan θ row _{90 °}, the one-dimensional vector X of structure " ₁, X " ₂x " _mfor:

$\left\{\begin{array}{c}{X}_1^{\′\′}=\left[\begin{array}{ccccc}{a}_{11}& a_{12}& ...& ...& a_{1m}\end{array}\right]\\ X_2^{\′\′}=\left[\begin{array}{ccccc}{a}_{21}& a_{22}& ...& ...& a_{2m}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m^{\′\′}=\left[\begin{array}{ccccc}{a}_{m1}& a_{m2}& ...& ...& a_{\mathrm{mm}}\end{array}\right]\end{array}\right.;$

3. by by after image array transposition and upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [0 °, 45 °] corresponding image array sub-block is built to one-dimensional vector; By by after image array upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [90 °, 135 °] corresponding image array sub-block is built to one-dimensional vector; By by after image array transposition, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [135 °, 180 °] corresponding image array sub-block is built to one-dimensional vector; Complete the structure of the one-dimensional vector to [0 °, 180 °] corresponding image array sub-block.

Described step 2) in, determine convolution algorithm result absolute value | f _σ(t) local maximum of * X|, it comprises the following steps: a 1. default maximum value threshold alpha, α=(0,1); 2. exist | f _σ(t) first crest place in * X| waveform, successively relatively before and after the size of data, when data transfer to while successively decreasing interval by increasing progressively interval, record this flex point value MAX|f _σ(t) * X|; Exist data to be less than α MAX|f when descending tops is interval _σ(t) when * X|, this crest effectively and its maximum value be MAX|f _σ(t) * X|; If not existing, this descending tops interval is less than α MAX|f _σ(t) value of * X| and proceed to the interval that increases progressively of next crest, this crest is invalid, continues to judge the maximum value of next crest; 3. non-maximum value is suppressed.

The present invention is owing to taking above technical scheme, it has the following advantages: 1, the present invention is due to according to [45 ° of rim detection angles, 90 °) tangent value, select a m capable × image subblock of mtan θ row cuts apart template image, by template image be divided into several m capable × image subblock of mtan θ row, then build some one-dimensional vector along the diagonal line of this image subblock, taking 45 °～90 ° direction rim detection as basis, by template image matrix is turned down and/or transposition after, realize the structure of one-dimensional vector to 0 °～180 ° of corresponding image array sub-blocks, thereby reduced the angular range that builds one-dimensional vector, by the one-dimensional vector of 0 °～180 ° of corresponding image array sub-blocks and Gaussian function first order derivative are carried out to convolution algorithm and convolution results are taken absolute value, make the angular range of edge direction detection from [0 °, 180 °] expand to [0 °, 360 °], therefore the mode that this one-dimensional vector that the present invention adopts is cut apart image has expanded the scope of rim detection angle, can realize the rim detection of any one angle direction that the line number m of selected image subblock is corresponding in angular interval, thereby can under the prerequisite that does not change image information, solve single pixel multi-angle rim detection problem.2, the present invention is due to the one-dimensional vector of structure is made to convolution algorithm with the first order derivative of Gaussian function respectively, and convolution algorithm result is taken absolute value, by finding the local maximum of each convolution algorithm result absolute value, make two dimensional image rim detection problem change one dimension waveform peak test problems into, therefore the present invention can reduce Image Edge-Detection difficulty.3, the present invention is owing to adopting default rim detection parameter to carry out rim detection to 18 kinds of given template images, by image border in the image border that relatively generates and template, obtain optimum rim detection parameter, and utilize this optimum rim detection parameter to carry out rim detection to pending image, the edge detection results that therefore adopts the present invention to obtain is more accurate.Based on above advantage, the present invention can be widely used in Image Edge-Detection.

Brief description of the drawings

Fig. 1 is the process flow diagram that the present invention is based on the multi-angle edge detection method of Gauss wavelet one dimension peak value identification;

Fig. 2 is 18 kinds of given template images; Wherein, figure (a) is the template image at 9 edges, species stage, and figure (b) is the template image at 9 kinds of line edges;

Fig. 3 utilizes the schematic diagram of image subblock to template image or pending Image Segmentation Using; Wherein, the border of " ... " presentation video block, "---" represents the border of template image or pending image, and θ represents rim detection angle, and 0 represents 0 pixel filling district;

Fig. 4 makes convolution algorithm, takes absolute value and peak value detects effect schematic diagram;

Fig. 5 is the schematic diagram of determining the local maximum of convolution algorithm result absolute value; Wherein, figure (a) is Valid peak schematic diagram, and figure (b) is invalid crest schematic diagram;

Fig. 6 is that 18 kinds of template images contrast schematic diagram with its edge detection results; Wherein, figure (a) is that the template image of step edges contrasts schematic diagram with its edge detection results, and figure (b) is that the template image at line edge contrasts schematic diagram with its edge detection results.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in detail.

As shown in Figure 1, the multi-angle edge detection method that the present invention is based on Gauss wavelet one dimension peak value identification comprises the following steps:

1) pending image is carried out to histogram analysis, obtain the gray-scale value of target and background thereof, adopt the gray-scale value obtaining to carry out assignment to 18 kinds of given template images, it specifically comprises:

(1) optional piece image is as pending image, and it is carried out to histogram analysis, obtains the gray-scale value of target in pending image and background thereof.

(2) given 18 kinds of template images as shown in Figure 2.

(3) utilize the gray-scale value that step (1) obtains respectively the target in 18 kinds of template images in step (2) and background thereof to be carried out to assignment.

2) default some groups of rim detection parameters: the first order derivative f of Gaussian function _σ(t) parameter σ, one dimension peak value maximum value threshold alpha and binary-state threshold, adopt some groups of default rim detection parameters respectively 18 kinds of template images after assignment to be carried out to rim detection, and find out the edge detection results the highest with the template image degree of approximation, will obtain this result rim detection parameter used as optimized parameter.

Only, to adopt one group of rim detection parameter to carry out rim detection as example to 18 kinds of template images, the process of finding optimized parameter is described, it specifically comprises the following steps:

(1) utilize image subblock to cut apart template image;

As shown in Figure 3, according to [45 ° of rim detection angle θ ∈, 90 °) tangent value, select a m capable × image subblock of mtan θ row adopt this image subblock to cut apart template image, by template image be divided into several m capable × image subblock of mtan θ row.When template image can not be divided into an integer m capable × when the image subblock of mtan θ row, template image is expanded, the expansion of template image is filled up with 0 pixel grey scale, make template image can be divided into an integer m capable × image subblock of mtan θ row.

(2) m capable × build some one-dimensional vector along the diagonal line of this image subblock in the image subblock of mtan θ row;

Owing to building in each image subblock, the method for one-dimensional vector is identical, therefore only taking a m wherein capable × image subblock of mtan θ row describes construction method as example, it specifically comprises the following steps:

1. according to the span of rim detection angle θ: θ ∈ [45 °, 90 °), provide m capable × three kinds of forms of the image subblock of mtan θ row and Q _{90 °}, and Q _{90 °}be expressed as with pixel a respectively:

θ ₁for the form of image subblock corresponding rim detection angle,

θ ₂for the form of image subblock corresponding rim detection angle,

with the rim detection angle corresponding to form of two kinds of image subblocks forms a closed interval [θ ₁, θ ₂].Its angular range is:

In formula, m > 1.

2. for m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ₁, X ₂x _mand Y ₁y _m-1,

$\left\{\begin{array}{c}{X}_1=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2k}& ...& a_{2(2k-1)},& ...& ...& a_{m[(m-1)(k-1)+1]}& ...& a_{m[m(k-1)+1]}\end{array}\right]\\ X_2=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3k}& ...& a_{3(2k-1)},& ...& ...& ,a_{m[(m-2)(k-1)+1]}& ...& a_{m[(m-1)(k-1)+1]}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}=\left[\begin{array}{ccccccccccc}{a}_{1k}& ...& a_{1(2k-1)},& a_{2(2k-1)}& ...& a_{2(3k-2)},& ...& ...& ,a_{(m-1)[(m-1)(k-1)+1]}& ...& a_{(m-1)[m(k-1)+1]}\end{array}\right]\\ Y_1=\left[\begin{array}{ccc}{a}_{1[(m-1)(k-1)+1]}& ...& a_{1[m(k-1)+1]}\end{array}\right]\end{array}\right.,$

For m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ' ₁, X' ₂x' _mand Y ' ₁y ' _m-1,

$\left\{\begin{array}{c}{X}_1^{\′}=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2(k+1)}& ...& a_{2\left(2k\right)},& ...& ...& a_{m\left[(m-1)(k-1\right]}& ...& a_{m\left(\mathrm{mk}\right)}\end{array}\right]\\ X_2^{\′}=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3(k+1)}& ...& a_{3\left(2k\right)},& ...& ...& ,a_{m[(m-2)k+1]}& ...& a_{m\left[(m-1)k\right]}\end{array}\right]\\ X_m^{\′}=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}^{\′}=\left[\begin{array}{ccccccccccc}{a}_{1(k+1)}& ...& a_{1\left(2k\right)},& a_{2(2k+1)}& ...& a_{2\left(3k\right)},& ...& ...& a_{(m-1)[(m-1)k+1]}& ...& a_{(m-1)\left(\mathrm{mk}\right)}\end{array}\right]\\ \·\\ \·\\ \·\\ Y_1^{\′}=\left[\begin{array}{ccc}{a}_{1[(m-1)k+1]}& ...& a_{1\left(\mathrm{mk}\right)}\end{array}\right]\end{array}\right..$

For m capable × the form Q of the image subblock of mtan θ row _{90 °}, the one-dimensional vector X of structure " ₁, X " ₂x " _mfor:

$\left\{\begin{array}{c}{X}_1^{\′\′}=\left[\begin{array}{ccccc}{a}_{11}& a_{12}& ...& ...& a_{1m}\end{array}\right]\\ X_2^{\′\′}=\left[\begin{array}{ccccc}{a}_{21}& a_{22}& ...& ...& a_{2m}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m^{\′\′}=\left[\begin{array}{ccccc}{a}_{m1}& a_{m2}& ...& ...& a_{\mathrm{mm}}\end{array}\right]\end{array}\right..$

3. by by after image array transposition and upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [0 °, 45 °] corresponding image array sub-block is built to one-dimensional vector.By by after image array upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [90 °, 135 °] corresponding image array sub-block is built to one-dimensional vector.By by after image array transposition, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [135 °, 180 °] corresponding image array sub-block is built to one-dimensional vector.Based on above method, according to the method that [45 °, 90 °] corresponding image array sub-block is built to one-dimensional vector, realize the structure of the one-dimensional vector to [0 °, 180 °] corresponding image array sub-block, thereby can reduce the angular range that builds one-dimensional vector.

(3) as shown in Figure 4, by the some one-dimensional vector X that build ₁, X ₂x _mand Y ₁y _m-1respectively with the first order derivative f of Gaussian function _σ(t) make convolution algorithm, and convolution algorithm result taken absolute value and obtained:

| f _σ(t) * X ₁|, | f _σ(t) * X ₂| ... | f _σ(t) * X _m| and | f _σ(t) * Y ₁| ... | f _σ(t) * Y _m-1| (4) thus the negative value that makes to occur in convolution algorithm result all change on the occasion of.

By the computing that the some one-dimensional vector that build are made convolution and taken absolute value, make rim detection angle expand to [0 °, 360 °] from [0 °, 180 °].Therefore only need process the one-dimensional vector of [0 °, 180 °] corresponding image array sub-block.

(4) by finding the absolute value of each convolution algorithm result | f _σ(t) * X| and | f _σ(t) local maximum of * Y|, makes two dimensional image rim detection problem change one dimension waveform peak test problems into.Wherein, X=X ₁, X ₂..., X _m, Y=Y ₁, Y ₂..., Y _m-1.

As shown in Figure 5, adopt following methods to determine the absolute value of convolution algorithm result | f _σ(t) local maximum of * X|, it specifically comprises the following steps:

1. a default maximum value threshold alpha, α=(0,1).

2. exist | f _σ(t) first crest place in * X| waveform, successively relatively before and after the size of data, when data transfer to while successively decreasing interval by increasing progressively interval, record this flex point value MAX|f _σ(t) * X|; Exist data to be less than α MAX|f when descending tops is interval _σ(t) when * X|, this crest effectively and its maximum value be MAX|f _σ(t) * X|; If not existing, this descending tops interval is less than α MAX|f _σ(t) value of * X| and proceed to the interval that increases progressively of next crest, this crest is invalid, continues to judge the maximum value of next crest.

3. non-maximum value is suppressed.

(5) maximum value step (4) being obtained is composed gray-scale value, and its value is:

According to pixel subscript, by the pixel in same position in the edge image pixel replacement original image generating.

(6) the some edge images that different edges detection angles direction obtained carry out gray scale stack.True edge position can brighten because of the repeatedly stack of pixel grey scale.According to the grey level histogram of image after repeatedly superposeing, using gray-scale value the highest gray scale number percent as binary-state threshold, according to this binary-state threshold, the edge of image after repeatedly superposeing is carried out to binary conversion treatment, generate final as shown in Figure 6 image border.

(7) adopt Pratt rim detection quality factor method that image border in the image border of generation and template image is compared, obtain with template image in the highest one group of edge detection results of the image border degree of approximation, and will obtain this edge detection results rim detection parameter used as optimized parameter.

3) adopt step 2) optimized parameter that obtains, repeating step 2) in step (1)～step (6), pending image is carried out to rim detection.

The various embodiments described above are only for illustrating the present invention; wherein the structure of each parts, connected mode and method step etc. all can change to some extent; every equivalents of carrying out on the basis of technical solution of the present invention and improvement, all should not get rid of outside protection scope of the present invention.

Claims (4)

1. the multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value, it comprises the following steps:

1) pending image is carried out to histogram analysis, obtain the gray-scale value of target and background thereof, adopt the gray-scale value obtaining to carry out assignment to 18 kinds of given template images;

2) default some groups of rim detection parameters, adopt some groups of default rim detection parameters respectively 18 kinds of template images after assignment to be carried out to rim detection, and find out the edge detection results the highest with the template image degree of approximation, to obtain this result rim detection parameter used as optimized parameter, it specifically comprises the following steps:

(1) utilize m capable × image subblock of mtan θ row cuts apart template image, by template image be divided into several m capable × image subblock of mtan θ row, wherein, θ is rim detection angle, θ ∈ [45 °, 90 °);

(2) m capable × build some one-dimensional vector along its diagonal line in the image subblock of mtan θ row;

(3) by the some one-dimensional vector that build respectively with the first order derivative f of Gaussian function _σ(t) make convolution algorithm, and convolution algorithm result is taken absolute value;

(4) determine the local maximum of convolution results absolute value;

(5) local maximum that obtains is composed to gray-scale value, and according to pixel subscript by the pixel in same position in the edge image pixel replacement original image generating;

(6) the some edge images that different edges detection angles direction obtained carry out gray scale stack, according to the grey level histogram of image after repeatedly superposeing, using gray-scale value the highest gray scale number percent as binary-state threshold, according to this binary-state threshold, the edge of image after repeatedly superposeing is carried out to binary conversion treatment;

(7) adopt Pratt rim detection quality factor method that image border in the image border of generation and template image is compared, obtain the one group edge detection results the highest with the template image edge degree of approximation, and will obtain this edge detection results rim detection parameter used as optimized parameter;

3) adopt step 2) optimized parameter that obtains, repeating step 2) in step (1)～step (6), pending image is carried out to rim detection.

2. a kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value as claimed in claim 1, is characterized in that: described step 2) in, default rim detection parameter comprises the first order derivative f of Gaussian function _σ(t) parameter σ, one dimension peak value maximum value threshold alpha and binary-state threshold.

3. a kind of multi-angle edge detection method based on Gauss wavelet one dimension peak value identification as claimed in claim 1 or 2, is characterized in that: described step 2) in, to m capable × image subblock of mtan θ row builds one-dimensional vector, it comprises:

1. according to the span of rim detection angle θ: θ ∈ [45 °, 90 °), provide m capable × image of mtan θ row

Three kinds of forms of piece and Q _{90 °}, and Q _{90 °}be expressed as with pixel a respectively:

θ ₁for the form of image subblock corresponding rim detection angle,

θ ₂for the form of image subblock corresponding rim detection angle,

2. for m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ₁, X ₂x _mand Y ₁y _m-1,

$\left\{\begin{array}{c}{X}_1=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2k}& ...& a_{2(2k-1)},& ...& ...& a_{m[(m-1)(k-1)+1]}& ...& a_{m[m(k-1)+1]}\end{array}\right]\\ X_2=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3k}& ...& a_{3(2k-1)},& ...& ...& ,a_{m[(m-2)(k-1)+1]}& ...& a_{m[(m-1)(k-1)+1]}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}=\left[\begin{array}{ccccccccccc}{a}_{1k}& ...& a_{1(2k-1)},& a_{2(2k-1)}& ...& a_{2(3k-2)},& ...& ...& ,a_{(m-1)[(m-1)(k-1)+1]}& ...& a_{(m-1)[m(k-1)+1]}\end{array}\right]\\ Y_1=\left[\begin{array}{ccc}{a}_{1[(m-1)(k-1)+1]}& ...& a_{1[m(k-1)+1]}\end{array}\right]\end{array}\right.,$

For m capable × form of the image subblock of mtan θ row according to following form

Along the form of image subblock diagonal get some one-dimensional vector X ' ₁, X' ₂x' _mand Y ' ₁y ' _m-1,

$\left\{\begin{array}{c}{X}_1^{\′}=\left[\begin{array}{ccccccccccc}{a}_{11}& ...& a_{1k},& a_{2(k+1)}& ...& a_{2\left(2k\right)},& ...& ...& a_{m\left[(m-1)(k-1\right]}& ...& a_{m\left(\mathrm{mk}\right)}\end{array}\right]\\ X_2^{\′}=\left[\begin{array}{ccccccccccc}{a}_{21}& ...& a_{2k},& a_{3(k+1)}& ...& a_{3\left(2k\right)},& ...& ...& ,a_{m[(m-2)k+1]}& ...& a_{m\left[(m-1)k\right]}\end{array}\right]\\ X_m^{\′}=\left[\begin{array}{ccc}{a}_{m1}& ...& a_{\mathrm{mk}}\end{array}\right]\\ Y_{m-1}^{\′}=\left[\begin{array}{ccccccccccc}{a}_{1(k+1)}& ...& a_{1\left(2k\right)},& a_{2(2k+1)}& ...& a_{2\left(3k\right)},& ...& ...& a_{(m-1)[(m-1)k+1]}& ...& a_{(m-1)\left(\mathrm{mk}\right)}\end{array}\right]\\ \·\\ \·\\ \·\\ Y_1^{\′}=\left[\begin{array}{ccc}{a}_{1[(m-1)k+1]}& ...& a_{1\left(\mathrm{mk}\right)}\end{array}\right]\end{array}\right.;$

For m capable × the form Q of the image subblock of mtan θ row _{90 °}, the one-dimensional vector X of structure " ₁, X " ₂x " _mfor:

$\left\{\begin{array}{c}{X}_1^{\′\′}=\left[\begin{array}{ccccc}{a}_{11}& a_{12}& ...& ...& a_{1m}\end{array}\right]\\ X_2^{\′\′}=\left[\begin{array}{ccccc}{a}_{21}& a_{22}& ...& ...& a_{2m}\end{array}\right]\\ \·\\ \·\\ \·\\ X_m^{\′\′}=\left[\begin{array}{ccccc}{a}_{m1}& a_{m2}& ...& ...& a_{\mathrm{mm}}\end{array}\right]\end{array}\right.;$

3. by by after image array transposition and upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [0 °, 45 °] corresponding image array sub-block is built to one-dimensional vector; By by after image array upset, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [90 °, 135 °] corresponding image array sub-block is built to one-dimensional vector; By by after image array transposition, adopt with [45 °, 90 °] corresponding image array sub-block and build the identical method of one-dimensional vector, [135 °, 180 °] corresponding image array sub-block is built to one-dimensional vector; Complete the structure of the one-dimensional vector to [0 °, 180 °] corresponding image array sub-block.

4. a kind of multi-angle edge detection method based on the identification of Gauss wavelet one dimension peak value as claimed in claim 1 or 2, is characterized in that: described step 2) in, determine convolution algorithm result absolute value | f _σ(t) local maximum of * X|, it comprises the following steps:

1. a default maximum value threshold alpha, α=(0,1);

2. exist | f _σ(t) first crest place in * X| waveform, successively relatively before and after the size of data, when data transfer to while successively decreasing interval by increasing progressively interval, record this flex point value MAX|f _σ(t) * X|; Exist data to be less than α MAX|f when descending tops is interval _σ(t) when * X|, this crest effectively and its maximum value be MAX|f _σ(t) * X|; If not existing, this descending tops interval is less than α MAX|f _σ(t) value of * X| and proceed to the interval that increases progressively of next crest, this crest is invalid, continues to judge the maximum value of next crest;

3. non-maximum value is suppressed.