CN109146902A - A kind of profile testing method based on color antagonism receptive field and monochrome channel - Google Patents
A kind of profile testing method based on color antagonism receptive field and monochrome channel Download PDFInfo
- Publication number
- CN109146902A CN109146902A CN201810876320.1A CN201810876320A CN109146902A CN 109146902 A CN109146902 A CN 109146902A CN 201810876320 A CN201810876320 A CN 201810876320A CN 109146902 A CN109146902 A CN 109146902A
- Authority
- CN
- China
- Prior art keywords
- channel
- pixel
- antagonism
- response
- double
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Processing Of Color Television Signals (AREA)
- Image Processing (AREA)
Abstract
The present invention is intended to provide the profile testing method based on color antagonism receptive field and monochrome channel, comprising the following steps: A, input image to be detected extract red, green, blue component, calculate yellow, black, white component;B, red green, blue yellow, 3 color combinations of black and white and corresponding channel are preset, respective single antagonism response is calculated;C, the multiple directions parameter for dividing equally circumference is preset, double antagonism receptive field filter templates of corresponding all directions parameter are constructed;Double antagonisms response in each channel of the pixel is calculated for each pixel;D, two double antagonisms for calculating each channel for each pixel inhibit profile response;E, for each pixel: after inhibiting profile response to be normalized double antagonisms of each channel whole, the maximum value of acquired results is the final profile response of the pixel.This method overcomes prior art defect, has the characteristics that detection effect is good, computational efficiency is high.
Description
Technical field
The present invention relates to image outline detection fields, and in particular to a kind of based on color antagonism receptive field and monochrome channel
Profile testing method.
Background technique
Edge detection is the necessary basis and premise of the image processing works such as target identification, image segmentation, accuracy pair
The application of object key feature and profile is depended on to have in compression of images, pattern-recognition, industrial detection, recognition of face etc. important
Influence.Deepening continuously and develop with image processing application, research, which finds profile more, can describe the shape of target in image
Information, contour detecting come into being.In the picture, profile and edge are different, those continuously embody the edge of subject goal
It is profile, and texture marginal information caused by those complex backgrounds is not then profile.Since most of images contain noise
Etc. background informations interference, efficiently extract the profile of main body, especially up to Detection accuracy is high, the requirement of registration is non-
It is often difficult.
Nineteen forty-six, Hungary scientist Gabor propose Gabor function to describe classical receptive field, simulate the court of receptive field
To response characteristics such as selection, band logicals.1962, Hubel etc. proposed that primary visual cortex neuron receptive field has towards selection
Characteristic, while finding that adjacent neurons receptive field has environmental stimuli similar towards selectivity in visual cortex.Nineteen sixty-five,
Rodieck has found that receptive field is in the structure of concentric circles antagonism, proposes classical receptive field model.1980, Marcelja was one-dimensional
Gabor function be generalized to two dimension, and point out two-dimensional Gabor function can very well simulation primary visual cortex simple cell it is classical
The response characteristic of receptive field.1992, Li Chaoyi had found the characteristic of non-classical receptive field, solved at image for visual processes mechanism
The problems such as reason, provides new thinking.2003, the inhibition using non-classical receptive field to classical receptive field such as Grigorescu
Characteristic realizes the inhibition of subject goal contour detecting and texture edge in image.Although the model improves contour detecting effect,
But a part of body profile can be curbed, to influence testing result;In addition, the model cannot accurately embody non-classical sense
By wild architectural characteristic.For this problem, 2007, Tang Qiling et al. proposed that a butterfly inhibits model, including lateral areas suppression
System and petiolarea are easily changed, and the integrality of contours extract is improved.In order to embody importance of the colouring information in contour detecting, 2013
Year, Yang Kaifu etc. proposes that CO model, the model extract the objective contour in image using color antagonism principle, achieves well
Detection effect, but the computational efficiency of the model is still up for improving.
Summary of the invention
The present invention is intended to provide a kind of profile testing method based on color antagonism receptive field and monochrome channel, this method gram
Prior art defect is taken, has the characteristics that detection effect is good, computational efficiency is high.
Technical scheme is as follows:
A kind of profile testing method based on color antagonism receptive field and monochrome channel, comprising the following steps:
A, image to be detected is inputted, the red, green, blue component extraction of each pixel in image to be detected is come out, and benefit
Calculate the yellow component, black component, white component of each pixel with red, green component, the value of the yellow component be red component with it is green
The half of component and value, the value of the black component are the minimum value of red, green, blue component, the value of the white component be it is red,
The maximum value of green, blue component or the value of the white component be red, green, blue component and value;
B, red green, blue yellow, three color combinations of black and white are preset, red green combination is equipped with red green channel, green/red channel, blue yellow
Combination is equipped with blue/yellow channel, yellow blue channel, and black-and-white group is equipped with black/white channel, channel white/black, calculate each pixel it is red/
Green channel, green/red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black;
C, the multiple directions parameter for dividing equally circumference is preset, double antagonism receptive field filters of corresponding all directions parameter are constructed
Template;
For each pixel, using each double antagonism receptive field filter templates to single antagonism response in each channel respectively into
Row filtering, obtains the boundary response value under all directions parameter in each channel of the pixel;For each channel, it is each to choose the channel
The maximum value of boundary response value under directioin parameter, double antagonisms as the channel respond;To obtain each logical of the pixel
Double antagonisms in road respond;
D, for each pixel, double antagonisms response in each channel is filtered using difference of Gaussian function, and to filtering
As a result it is normalized to obtain double antagonism inhibition responses in each channel;Double antagonisms response in each channel subtracts the double of the channel
Antagonism inhibition response obtains first pair of antagonism and inhibits profile response, and double antagonisms response in each channel subtracts color belonging to the channel
Double antagonism inhibition responses in another channel in combination obtain second pair of antagonism and inhibit profile response;
E, for each pixel: by all first pair of antagonisms inhibit profile response and the second pair of antagonism inhibit profile respond into
After row normalization, the maximum value of acquired results is the final profile response of the pixel.
Preferably, in the step each channel single antagonism response computation formula are as follows:
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (1);
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (2);
Sby(x, y)=ω1·B(x,y)+ω2·Y(x,y) (3);
Syb(x, y)=ω1·Y(x,y)+ω2·B(x,y) (4);
Sblw(x, y)=ω1·BL(x,y)+ω2·W(x,y) (5);
Swbl(x, y)=ω1·W(x,y)+ω2·BL(x,y) (6);
Wherein, the red component of R (x, y) expression pixel (x, y), the green component of G (x, y) expression pixel (x, y), B (x,
Y) the blue component of pixel (x, y) is indicated;Y (x, y) indicates the yellow component of pixel (x, y),
BL (x, y) indicates that the black component of pixel (x, y), BL (x, y)=min (R (x, y), G (x, y), B (x, y)), W (x, y) indicate
The white component of pixel (x, y), W (x, y)=max (R (x, y), G (x, y), B (x, y)) or W (x, y)=R (x, y)+G (x,
y)+B(x,y);
Srg(x,y)、Srg(x,y)、Sby(x,y)、Syb(x,y)、Sblw(x,y)、Swbl(x, y) be respectively red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black, wherein
Preferably, the boundary response value calculation formula under all directions parameter in each channel of each pixel of step C
It is as follows:
Wherein rg, gr, by, yb, blw, wbl respectively indicate red green channel, green/red channel, indigo plant/Huang channel, yellow blue logical
Road, black/white channel, channel white/black;θiFor directioin parameter, Crg、Cgr、Cby、Cyb、Cblw、CwblRespectively indicate red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the double antagonism receptive field filter template ranges in channel white/black, m, n
The offset of the horizontal axis of respectively double antagonism receptive field filter templates, the longitudinal axis;NθFor side
To the number of parameter;
RF(m,n;θi) it is double antagonism receptive field filter functions;Wherein:
σ1For the dimensional parameters of double antagonism receptive fields;
Double antagonism response computation formula in each channel are as follows:
Drg(x, y)=max { Drg(x,y;θi) | i=1,2 ... Nθ} (16);
Dgr(x, y)=max { Dgr(x,y;θi) | i=1,2 ... Nθ} (17);
Dby(x, y)=max { Dby(x,y;θi) | i=1,2 ... Nθ} (18);
Dyb(x, y)=max { Dyb(x,y;θi) | i=1,2 ... Nθ} (19);
Dblw(x, y)=max { Dblw(x,y;θi) | i=1,2 ... Nθ} (20);
Dwbl(x, y)=max { Dwbl(x,y;θi) | i=1,2 ... Nθ} (21)。
Preferably, in the step D, difference of Gaussian function is as follows:
Wherein, σ2For the bandwidth of difference of Gaussian function template central area;
Filter result is as follows:
DoGab(x,y;σ2)=Dab(x,y)*DoG(x,y;σ2) (23);Wherein ab=(rg, gr, by, yb, blw,
wbl);
The normalized function are as follows:
The double antagonisms of two of each each channel of pixel inhibit profile response computation formula as follows:
R1 rg(x,y;σ2)=Drg(x,y)-αωrg(x,y;σ2) (25);
R2 rg(x,y;σ2)=Drg(x,y)-αωgr(x,y;σ2) (26);
R1 gr(x,y;σ2)=Dgr(x,y)-αωgr(x,y;σ2) (27);
R2 gr(x,y;σ2)=Dgr(x,y)-αωrg(x,y;σ2) (28);
R1 by(x,y;σ2)=Dby(x,y)-αωby(x,y;σ2) (29);
R2 by(x,y;σ2)=Dby(x,y)-αωyb(x,y;σ2) (30);
R1 yb(x,y;σ2)=Dyb(x,y)-αωyb(x,y;σ2) (31);
R2 yb(x,y;σ2)=Dyb(x,y)-αωby(x,y;σ2) (32);
R1 blw(x,y;σ2)=Dblw(x,y)-αωblw(x,y;σ2) (33);
R2 blw(x,y;σ2)=Dblw(x,y)-αωwbl(x,y;σ2) (34);
R1 wbl(x,y;σ2)=Dwbl(x,y)-αωwbl(x,y;σ2) (35);
R2 wbl(x,y;σ2)=Dwbl(x,y)-αωblw(x,y;σ2) (36);
Wherein R1 rg(x,y;σ2)、R2 rg(x,y;σ2) responded for two double antagonisms inhibition profiles in red green channel.
Preferably, in the step E, the normalization formula for inhibiting profile response to carry out each double antagonisms is as follows:
Wherein XnIndicate the normalized value of data X, Xmax、XminRespectively indicate the maximum value and minimum value in data X;
T (x, y)=max (R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln,
R2 wbln);
The final profile that wherein T (x, y) is pixel (x, y) responds:
R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln, R2 wblnRespectively each pair
Antagonism inhibits the normalized value of profile response.
The present invention is carried out using directioin parameter with double antagonisms inhibition towards selectivity, and is carried out in conjunction with DOG function
Texture inhibits, and simulation peripheral nerve member cell receptive field region, which generates the response of central nervous member, changes antagonism with distance;
Meanwhile double antagonisms that black-white colors channel further increases Detection accuracy, and considers Color Channel itself are increased,
Therefore double antagonisms are carried out to same color channel and inhibits profile response computation, improve the effect of texture inhibition, improve profile inspection
Survey success rate.
Detailed description of the invention
Fig. 1 is the contour detecting effect contrast figure that the embodiment of the present invention 1 provides.
Fig. 2 is the contour detecting effect contrast figure that the embodiment of the present invention 2 provides.
Specific embodiment
The present invention is illustrated with reference to the accompanying drawings and examples.
Embodiment 1
Profile testing method provided in this embodiment based on color antagonism receptive field and monochrome channel, including following step
It is rapid:
A, image to be detected is inputted, the red, green, blue component extraction of each pixel in image to be detected is come out, and benefit
Calculate the yellow component, black component, white component of each pixel with red, green component, the value of the yellow component be red component with it is green
The half of component and value, the value of the black component are the minimum value of red, green, blue component, the value of the white component be it is red,
The maximum value of green, blue component;
B, red green, blue yellow, three color combinations of black and white are preset, red green combination is equipped with red green channel, green/red channel, blue yellow
Combination is equipped with blue/yellow channel, yellow blue channel, and black-and-white group is equipped with black/white channel, channel white/black, calculate each pixel it is red/
Green channel, green/red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black;
Single antagonism response computation formula in each channel in the step B are as follows:
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (1);
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (2);
Sby(x, y)=ω1·B(x,y)+ω2·Y(x,y) (3);
Syb(x, y)=ω1·Y(x,y)+ω2·B(x,y) (4);
Sblw(x, y)=ω1·BL(x,y)+ω2·W(x,y) (5);
Swbl(x, y)=ω1·W(x,y)+ω2·BL(x,y) (6);
Wherein, the red component of R (x, y) expression pixel (x, y), the green component of G (x, y) expression pixel (x, y), B (x,
Y) the blue component of pixel (x, y) is indicated;Y (x, y) indicates the yellow component of pixel (x, y),
BL (x, y) indicates that the black component of pixel (x, y), BL (x, y)=min (R (x, y), G (x, y), B (x, y)), W (x, y) indicate
The white component of pixel (x, y), W (x, y)=max (R (x, y), G (x, y), B (x, y));
Srg(x,y)、Srg(x,y)、Sby(x,y)、Syb(x,y)、Sblw(x,y)、Swbl(x, y) be respectively red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black, wherein
C, 12 directioin parameters for dividing equally circumference are preset, double antagonism receptive field filters of corresponding all directions parameter are constructed
Template;
For each pixel, using each double antagonism receptive field filter templates to single antagonism response in each channel respectively into
Row filtering, obtains the boundary response value under all directions parameter in each channel of the pixel;For each channel, it is each to choose the channel
The maximum value of boundary response value under directioin parameter, double antagonisms as the channel respond;To obtain each logical of the pixel
Double antagonisms in road respond;
Boundary response value calculation formula in the step C under all directions parameter in each channel of each pixel is as follows:
Boundary response value under all directions parameter in each channel of each pixel is as follows
Wherein rg, gr, by, yb, blw, wbl respectively indicate red green channel, green/red channel, indigo plant/Huang channel, yellow blue logical
Road, black/white channel, channel white/black;θiFor directioin parameter, Crg、Cgr、Cby、Cyb、Cblw、CwblRespectively indicate red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the double antagonism receptive field filter template ranges in channel white/black, m, n
The offset of the horizontal axis of respectively double antagonism receptive field filter templates, the longitudinal axis;RF(m,n;
θi) it is double antagonism receptive field filter functions;Wherein:
σ1For the dimensional parameters of double antagonism receptive fields;
Double antagonism response computation formula in each channel are as follows:
Drg(x, y)=max { Drg(x,y;θi) | i=1,2 ... 12 } (16);
Dgr(x, y)=max { Dgr(x,y;θi) | i=1,2 ... 12 } (17);
Dby(x, y)=max { Dby(x,y;θi) | i=1,2 ... 12 } (18);
Dyb(x, y)=max { Dyb(x,y;θi) | i=1,2 ... 12 } (19);
Dblw(x, y)=max { Dblw(x,y;θi) | i=1,2 ... 12 } (20);
Dwbl(x, y)=max { Dwbl(x,y;θi) | i=1,2 ... 12 } (21);
D, for each pixel, double antagonisms response in each channel is filtered using difference of Gaussian function, and to filtering
As a result it is normalized to obtain double antagonism inhibition responses in each channel;Double antagonisms response in each channel subtracts the double of the channel
Antagonism inhibition response obtains first pair of antagonism and inhibits profile response, and double antagonisms response in each channel subtracts color belonging to the channel
Double antagonism inhibition responses in another channel in combination obtain second pair of antagonism and inhibit profile response;
In the step D, difference of Gaussian function is as follows:
Wherein, σ2For the bandwidth of difference of Gaussian function template central area;
Filter result is as follows:
DoGab(x,y;σ2)=Dab(x,y)*DoG(x,y;σ2) (23);
Wherein ab=(rg, gr, by, yb, blw, wbl);
The normalized function are as follows:
The double antagonisms of two of each each channel of pixel inhibit profile response as follows:
R1 rg(x,y;σ2)=Drg(x,y)-αωrg(x,y;σ2) (25);
R2 rg(x,y;σ2)=Drg(x,y)-αωgr(x,y;σ2) (26);
R1 gr(x,y;σ2)=Dgr(x,y)-αωgr(x,y;σ2) (27);
R2 gr(x,y;σ2)=Dgr(x,y)-αωrg(x,y;σ2) (28);
R1 by(x,y;σ2)=Dby(x,y)-αωby(x,y;σ2) (29);
R2 by(x,y;σ2)=Dby(x,y)-αωyb(x,y;σ2) (30);
R1 yb(x,y;σ2)=Dyb(x,y)-αωyb(x,y;σ2) (31);
R2 yb(x,y;σ2)=Dyb(x,y)-αωby(x,y;σ2) (32);
R1 blw(x,y;σ2)=Dblw(x,y)-αωblw(x,y;σ2) (33);
R2 blw(x,y;σ2)=Dblw(x,y)-αωwbl(x,y;σ2) (34);
R1 wbl(x,y;σ2)=Dwbl(x,y)-αωwbl(x,y;σ2) (35);
R2 wbl(x,y;σ2)=Dwbl(x,y)-αωblw(x,y;σ2) (36);
Wherein R1 rg(x,y;σ2)、R2 rg(x,y;σ2) responded for two double antagonisms inhibition profiles in red green channel;
E, for each pixel: by all first pair of antagonisms inhibit profile response and the second pair of antagonism inhibit profile respond into
After row normalization, the maximum value of acquired results is the final profile response of the pixel;
In the step E, the normalization formula for inhibiting profile response to carry out each double antagonisms is as follows:
Wherein XnIndicate the normalized value of data X, Xmax、XminRespectively indicate the maximum value and minimum value in data X;
T (x, y)=max (R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln,
R2 wbln);
The final profile that wherein T (x, y) is pixel (x, y) responds:
R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln, R2 wblnRespectively each pair
Antagonism inhibits the normalized value of profile response.
The contour detecting isotropic model for below providing the profile testing method of the present embodiment and document 1 carries out effective
Property comparison, document 1 is as follows:
Document 1:Yang, K., et al.Efficient Color Boundary Detection with Color-
Opponent Mechanisms.in IEEE Conference on Computer Vision and Pattern
Recognition.2013;
Wherein Performance Evaluating Indexes P uses following standard:
N in formulaTP、nFP、nFNThe number of the profile of correct profile, error profile and omission that detection obtains is respectively indicated,
Evaluation metrics P value indicates that the effect of contour detecting is better between [0,1], closer to 1, in addition, defining tolerance are as follows: in 5*
The all calculations detected in 5 neighborhood correctly detect;
It chooses the 4 width pictures from BSDS300 database and carries out Usefulness Pair ratio, picture number is as follows: 302008 (pictures
1), 376043 (pictures 2), 119082 (pictures 3), 159008 (pictures 4);1 side of method and embodiment in document 1 is respectively adopted
Method carries out contour detecting to above-mentioned 4 width figure, and wherein the parameter group of 1 method selection of embodiment is as shown in table 1,1 method selection of document
Parameter group it is as shown in table 2;
1 embodiment of table, 1 parameter group table
2 document of table, 1 parameter group table
The optimal profile figure detected as shown in Figure 1 for the respectively original image of picture 1-4, actual profile figure, 1 method of document,
The optimal profile figure of 1 method of embodiment detection;The optimal P value detected as shown in table 3 for 1 method of document of above-mentioned 4 width image
With the optimal P value of 1 method of embodiment detection;
The comparison of table 3P value
No matter it can be seen from the results above that implementing from the effect of contours extract or from performance indicator parameter
1 method of example is superior to the method in document 1.
Embodiment 2
Profile testing method provided in this embodiment based on color antagonism receptive field and monochrome channel, including following step
It is rapid:
A, image to be detected is inputted, the red, green, blue component extraction of each pixel in image to be detected is come out, and benefit
Calculate the yellow component, black component, white component of each pixel with red, green component, the value of the yellow component be red component with it is green
The half of component and value, the value of the black component are the minimum value of red, green, blue component, the value of the white component be it is red,
Green, blue component and value;
B, red green, blue yellow, three color combinations of black and white are preset, red green combination is equipped with red green channel, green/red channel, blue yellow
Combination is equipped with blue/yellow channel, yellow blue channel, and black-and-white group is equipped with black/white channel, channel white/black, calculate each pixel it is red/
Green channel, green/red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black;
Single antagonism response computation formula in each channel in the step B are as follows:
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (1);
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (2);
Sby(x, y)=ω1·B(x,y)+ω2·Y(x,y) (3);
Syb(x, y)=ω1·Y(x,y)+ω2·B(x,y) (4);
Sblw(x, y)=ω1·BL(x,y)+ω2·W(x,y) (5);
Swbl(x, y)=ω1·W(x,y)+ω2·BL(x,y) (6);
Wherein, the red component of R (x, y) expression pixel (x, y), the green component of G (x, y) expression pixel (x, y), B (x,
Y) the blue component of pixel (x, y) is indicated;Y (x, y) indicates the yellow component of pixel (x, y),
BL (x, y) indicates that the black component of pixel (x, y), BL (x, y)=min (R (x, y), G (x, y), B (x, y)), W (x, y) indicate
The white component of pixel (x, y), W (x, y)=R (x, y)+G (x, y)+B (x, y);
Srg(x,y)、Srg(x,y)、Sby(x,y)、Syb(x,y)、Sblw(x,y)、Swbl(x, y) be respectively red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black, wherein
C, 12 directioin parameters for dividing equally circumference are preset, double antagonism receptive field filters of corresponding all directions parameter are constructed
Template;
For each pixel, using each double antagonism receptive field filter templates to single antagonism response in each channel respectively into
Row filtering, obtains the boundary response value under all directions parameter in each channel of the pixel;For each channel, it is each to choose the channel
The maximum value of boundary response value under directioin parameter, double antagonisms as the channel respond;To obtain each logical of the pixel
Double antagonisms in road respond;
In the step C, the boundary response value calculation formula under all directions parameter in each channel of each pixel is as follows:
Wherein rg, gr, by, yb, blw, wbl respectively indicate red green channel, green/red channel, indigo plant/Huang channel, yellow blue logical
Road, black/white channel, channel white/black;θiFor directioin parameter, Crg、Cgr、Cby、Cyb、Cblw、CwblRespectively indicate red green channel, it is green/
Red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the double antagonism receptive field filter template ranges in channel white/black, m, n
The offset of the horizontal axis of respectively double antagonism receptive field filter templates, the longitudinal axis;RF(m,n;
θi) it is double antagonism receptive field filter functions;Wherein:
σ1For the dimensional parameters of double antagonism receptive fields;
Double antagonism response computation formula in each channel are as follows:
Drg(x, y)=max { Drg(x,y;θi) | i=1,2 ... 12 } (16);
Dgr(x, y)=max { Dgr(x,y;θi) | i=1,2 ... 12 } (17);
Dby(x, y)=max { Dby(x,y;θi) | i=1,2 ... 12 } (18);
Dyb(x, y)=max { Dyb(x,y;θi) | i=1,2 ... 12 } (19);
Dblw(x, y)=max { Dblw(x,y;θi) | i=1,2 ... 12 } (20);
Dwbl(x, y)=max { Dwbl(x,y;θi) | i=1,2 ... 12 } (21);
D, for each pixel, double antagonisms response in each channel is filtered using difference of Gaussian function, and to filtering
As a result it is normalized to obtain double antagonism inhibition responses in each channel;Double antagonisms response in each channel subtracts the double of the channel
Antagonism inhibition response obtains first pair of antagonism and inhibits profile response, and double antagonisms response in each channel subtracts color belonging to the channel
Double antagonism inhibition responses in another channel in combination obtain second pair of antagonism and inhibit profile response;
In the step D, difference of Gaussian function is as follows:
Wherein, σ2For the bandwidth of difference of Gaussian function template central area;
Filter result is as follows:
DoGab(x,y;σ2)=Dab(x,y)*DoG(x,y;σ2) (23);
Wherein ab=(rg, gr, by, yb, blw, wbl);
The normalized function are as follows:
The double antagonisms of two of each each channel of pixel inhibit profile response as follows:
R1 rg(x,y;σ2)=Drg(x,y)-αωrg(x,y;σ2) (25);
R2 rg(x,y;σ2)=Drg(x,y)-αωgr(x,y;σ2) (26);
R1 gr(x,y;σ2)=Dgr(x,y)-αωgr(x,y;σ2) (27);
R2 gr(x,y;σ2)=Dgr(x,y)-αωrg(x,y;σ2) (28);
R1 by(x,y;σ2)=Dby(x,y)-αωby(x,y;σ2) (29);
R2 by(x,y;σ2)=Dby(x,y)-αωyb(x,y;σ2) (30);
R1 yb(x,y;σ2)=Dyb(x,y)-αωyb(x,y;σ2) (31);
R2 yb(x,y;σ2)=Dyb(x,y)-αωby(x,y;σ2) (32);
R1 blw(x,y;σ2)=Dblw(x,y)-αωblw(x,y;σ2) (33);
R2 blw(x,y;σ2)=Dblw(x,y)-αωwbl(x,y;σ2) (34);
R1 wbl(x,y;σ2)=Dwbl(x,y)-αωwbl(x,y;σ2) (35);
R2 wbl(x,y;σ2)=Dwbl(x,y)-αωblw(x,y;σ2) (36);
Wherein R1 rg(x,y;σ2)、R2 rg(x,y;σ2) responded for two double antagonisms inhibition profiles in red green channel;
E, for each pixel: by all first pair of antagonisms inhibit profile response and the second pair of antagonism inhibit profile respond into
After row normalization, the maximum value of acquired results is the final profile response of the pixel;
In the step E, the normalization formula for inhibiting profile response to carry out each double antagonisms is as follows:
Wherein XnIndicate the normalized value of data X, Xmax、XminRespectively indicate the maximum value and minimum value in data X;
T (x, y)=max (R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln,
R2 wbln);
The final profile that wherein T (x, y) is pixel (x, y) responds:
R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln, R2 wblnRespectively each pair
Antagonism inhibits the normalized value of profile response.
The contour detecting isotropic model for below providing the profile testing method of the present embodiment and document 1 carries out effective
Property comparison, document 1 is as follows:
Document 1:Yang, K., et al.Efficient Color Boundary Detection with Color-
Opponent Mechanisms.in IEEE Conference on Computer Vision and Pattern
Recognition.2013;
Wherein Performance Evaluating Indexes P uses following standard:
N in formulaTP、nFP、nFNThe number of the profile of correct profile, error profile and omission that detection obtains is respectively indicated,
Evaluation metrics P value indicates that the effect of contour detecting is better between [0,1], closer to 1, in addition, defining tolerance are as follows: in 5*
The all calculations detected in 5 neighborhood correctly detect;
It chooses the 4 width pictures from BSDS300 database and carries out Usefulness Pair ratio, picture number is as follows: 302008 (pictures
1), 376043 (pictures 2), 119082 (pictures 3), 159008 (pictures 4);2 side of method and embodiment in document 1 is respectively adopted
Method carries out contour detecting to above-mentioned 4 width figure, and wherein the parameter group of 2 method selection of embodiment is as shown in table 4,1 method selection of document
Parameter group it is as shown in table 5;
4 embodiment of table, 2 parameter group table
5 document of table, 1 parameter group table
It is illustrated in figure 2 the optimal profile figure that respectively original image of picture 1-4, actual profile figure, 1 method of document detect,
The optimal profile figure of 2 method of embodiment detection;The optimal P value detected as shown in table 6 for 1 method of document of above-mentioned 4 width image
With the optimal P value of 2 method of embodiment detection;
The comparison of 6 P value of table
No matter it can be seen from the results above that implementing from the effect of contours extract or from performance indicator parameter
2 method of example is superior to the method in document 1.
Claims (5)
1. a kind of profile testing method based on color antagonism receptive field and monochrome channel, it is characterised in that the following steps are included:
A, input image to be detected, the red, green, blue component extraction of each pixel in image to be detected is come out, and using it is red,
Green component calculates the yellow component, black component, white component of each pixel, the value of the yellow component be red component and green component and
The half of value, the value of the black component are the minimum value of red, green, blue component, and the value of the white component is red, green, blue point
The maximum value of amount;
B, red green, blue yellow, three color combinations of black and white are preset, red green combination is equipped with red green channel, green/red channel, blue yellow combination
Equipped with blue/yellow channel, yellow blue channel, black-and-white group is equipped with black/white channel, channel white/black, calculates the red green logical of each pixel
Road, green/red channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black;
C, the multiple directions parameter for dividing equally circumference is preset, double antagonism receptive field filter moulds of corresponding all directions parameter are constructed
Plate;
For each pixel, single antagonism response in each channel is filtered respectively using each double antagonism receptive field filter templates
Wave obtains the boundary response value under all directions parameter in each channel of the pixel;For each channel, the channel all directions are chosen
The maximum value of boundary response value under parameter, double antagonisms as the channel respond;Thus obtain each channel of the pixel
Double antagonism responses;
D, for each pixel, double antagonisms response in each channel is filtered using difference of Gaussian function, and to filter result
It is normalized to obtain double antagonism inhibition responses in each channel;Double antagonisms response in each channel subtracts double antagonisms in the channel
Inhibition response obtains first pair of antagonism and inhibits profile response, and double antagonisms response in each channel subtracts the combination of color belonging to the channel
In double antagonism inhibition responses in another channel obtain the second pair of antagonism and inhibit profile response;
E, for each pixel: inhibiting profile response and second pair of antagonism that profile response is inhibited to return all first pair of antagonisms
After one changes, the maximum value of acquired results is the final profile response of the pixel.
2. the profile testing method based on color antagonism receptive field and monochrome channel as described in claim 1, it is characterised in that:
Single antagonism response computation formula in each channel in the step B are as follows:
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (1);
Srg(x, y)=ω1·R(x,y)+ω2·G(x,y) (2);
Sby(x, y)=ω1·B(x,y)+ω2·Y(x,y) (3);
Syb(x, y)=ω1·Y(x,y)+ω2·B(x,y) (4);
Sblw(x, y)=ω1·BL(x,y)+ω2·W(x,y) (5);
Swbl(x, y)=ω1·W(x,y)+ω2·BL(x,y) (6);
Wherein, R (x, y) indicates that the red component of pixel (x, y), G (x, y) indicate the green component of pixel (x, y), B (x, y) table
Show the blue component of pixel (x, y);Y (x, y) indicates the yellow component of pixel (x, y),BL
(x, y) indicates that the black component of pixel (x, y), BL (x, y)=min (R (x, y), G (x, y), B (x, y)), W (x, y) indicate picture
The white component of vegetarian refreshments (x, y), W (x, y)=max (R (x, y), G (x, y), B (x, y));
Srg(x,y)、Srg(x,y)、Sby(x,y)、Syb(x,y)、Sblw(x,y)、Swbl(x, y) be respectively red green channel, green/red logical
Road, indigo plant/Huang channel, yellow blue channel, black/white channel, the respective single antagonism response in channel white/black, wherein
3. the profile testing method based on color antagonism receptive field and monochrome channel as described in claim 1, it is characterised in that:
Boundary response value calculation formula in the step C under all directions parameter in each channel of each pixel is as follows:
Wherein rg, gr, by, yb, blw, wbl respectively indicate red green channel, green/red channel, indigo plant/Huang channel, yellow blue channel,
Black/white channel, channel white/black;θiFor directioin parameter, Crg、Cgr、Cby、Cyb、Cblw、CwblRespectively indicate red green channel, green/red
Channel, indigo plant/Huang channel, yellow blue channel, black/white channel, the double antagonism receptive field filter template ranges in channel white/black, m, n points
Not Wei the horizontal axis of double antagonism receptive field filter templates, the longitudinal axis offset;
NθFor the number of directioin parameter;
RF(m,n;θi) it is double antagonism receptive field filter functions;Wherein:
σ1For the dimensional parameters of double antagonism receptive fields;
Double antagonism response computation formula in each channel are as follows:
Drg(x, y)=max { Drg(x,y;θi) | i=1,2 ... Nθ} (16);
Dgr(x, y)=max { Dgr(x,y;θi) | i=1,2 ... Nθ} (17);
Dby(x, y)=max { Dby(x,y;θi) | i=1,2 ... Nθ} (18);
Dyb(x, y)=max { Dyb(x,y;θi) | i=1,2 ... Nθ} (19);
Dblw(x, y)=max { Dblw(x,y;θi) | i=1,2 ... Nθ} (20);
Dwbl(x, y)=max { Dwbl(x,y;θi) | i=1,2 ... Nθ} (21)。
4. the profile testing method based on color antagonism receptive field and monochrome channel as described in claim 1, it is characterised in that:
In the step D, difference of Gaussian function is as follows:
Wherein, σ2For the bandwidth of difference of Gaussian function template central area;
Filter result is as follows:
DoGab(x,y;σ2)=Dab(x,y)*DoG(x,y;σ2) (23);
Wherein ab=(rg, gr, by, yb, blw, wbl);
The normalized function are as follows:
The double antagonisms of two of each each channel of pixel inhibit profile response computation formula as follows:
R1 rg(x,y;σ2)=Drg(x,y)-αωrg(x,y;σ2) (25);
R2 rg(x,y;σ2)=Drg(x,y)-αωgr(x,y;σ2) (26);
R1 gr(x,y;σ2)=Dgr(x,y)-αωgr(x,y;σ2) (27);
R2 gr(x,y;σ2)=Dgr(x,y)-αωrg(x,y;σ2) (28);
R1 by(x,y;σ2)=Dby(x,y)-αωby(x,y;σ2) (29);
R2 by(x,y;σ2)=Dby(x,y)-αωyb(x,y;σ2) (30);
R1 yb(x,y;σ2)=Dyb(x,y)-αωyb(x,y;σ2) (31);
R2 yb(x,y;σ2)=Dyb(x,y)-αωby(x,y;σ2) (32);
R1 blw(x,y;σ2)=Dblw(x,y)-αωblw(x,y;σ2) (33);
R2 blw(x,y;σ2)=Dblw(x,y)-αωwbl(x,y;σ2) (34);
R1 wbl(x,y;σ2)=Dwbl(x,y)-αωwbl(x,y;σ2) (35);
R2 wbl(x,y;σ2)=Dwbl(x,y)-αωblw(x,y;σ2) (36);
Wherein R1 rg(x,y;σ2)、R2 rg(x,y;σ2) responded for two double antagonisms inhibition profiles in red green channel.
5. the profile testing method based on color antagonism receptive field and monochrome channel as described in claim 1, it is characterised in that:
In the step E, the normalization formula for inhibiting profile response to carry out each double antagonisms is as follows:
Wherein XnIndicate the normalized value of data X, Xmax、XminRespectively indicate the maximum value and minimum value in data X;
T (x, y)=max (R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln, R2 wbln);
The final profile that wherein T (x, y) is pixel (x, y) responds:
R1 rgn, R2 rgn,R1 grn, R2 grn,R1 ybn, R2 ybn、R1 byn, R2 byn、R1 blwn, R2 blwn、R1 wbln, R2 wblnRespectively each double antagonism suppressions
The normalized value of ratch exterior feature response.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810876320.1A CN109146902B (en) | 2018-08-03 | 2018-08-03 | Contour detection method based on color antagonistic receptive field and black and white channel |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810876320.1A CN109146902B (en) | 2018-08-03 | 2018-08-03 | Contour detection method based on color antagonistic receptive field and black and white channel |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109146902A true CN109146902A (en) | 2019-01-04 |
CN109146902B CN109146902B (en) | 2021-09-21 |
Family
ID=64791454
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810876320.1A Active CN109146902B (en) | 2018-08-03 | 2018-08-03 | Contour detection method based on color antagonistic receptive field and black and white channel |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109146902B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040175057A1 (en) * | 2003-03-04 | 2004-09-09 | Thomas Tsao | Affine transformation analysis system and method for image matching |
CN106033608A (en) * | 2015-07-24 | 2016-10-19 | 广西科技大学 | Target contour detection method of biomimetic smooth tracking eye movement information processing mechanism |
CN106228547A (en) * | 2016-07-15 | 2016-12-14 | 华中科技大学 | A kind of view-based access control model color theory and homogeneity suppression profile and border detection algorithm |
-
2018
- 2018-08-03 CN CN201810876320.1A patent/CN109146902B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040175057A1 (en) * | 2003-03-04 | 2004-09-09 | Thomas Tsao | Affine transformation analysis system and method for image matching |
CN106033608A (en) * | 2015-07-24 | 2016-10-19 | 广西科技大学 | Target contour detection method of biomimetic smooth tracking eye movement information processing mechanism |
CN106228547A (en) * | 2016-07-15 | 2016-12-14 | 华中科技大学 | A kind of view-based access control model color theory and homogeneity suppression profile and border detection algorithm |
Non-Patent Citations (1)
Title |
---|
KAIFU YANG ET AL.: "Efficient Color Boundary Detection with Color-Opponent Mechanisms", 《2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Also Published As
Publication number | Publication date |
---|---|
CN109146902B (en) | 2021-09-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109146901A (en) | Profile testing method based on color antagonism receptive field | |
CN106228547B (en) | A kind of profile and border detection algorithm of view-based access control model color theory and homogeneity inhibition | |
Li et al. | Multi-angle head pose classification when wearing the mask for face recognition under the COVID-19 coronavirus epidemic | |
CN104318548B (en) | Rapid image registration implementation method based on space sparsity and SIFT feature extraction | |
WO2019104767A1 (en) | Fabric defect detection method based on deep convolutional neural network and visual saliency | |
CN105005766B (en) | A kind of body color recognition methods | |
Yan et al. | A skeletal similarity metric for quality evaluation of retinal vessel segmentation | |
CN104408711B (en) | Multi-scale region fusion-based salient region detection method | |
CN106778517A (en) | A kind of monitor video sequence image vehicle knows method for distinguishing again | |
CN105023253A (en) | Visual underlying feature-based image enhancement method | |
CN113793337B (en) | Locomotive accessory surface abnormal degree evaluation method based on artificial intelligence | |
Wang et al. | Skin color detection under complex background | |
CN104408728A (en) | Method for detecting forged images based on noise estimation | |
CN107123130A (en) | Kernel correlation filtering target tracking method based on superpixel and hybrid hash | |
CN107103266A (en) | The training of two-dimension human face fraud detection grader and face fraud detection method | |
Cheng et al. | Superpixel classification for initialization in model based optic disc segmentation | |
CN109087324A (en) | Profile testing method based on color antagonism receptive field and monochrome channel | |
CN108960276B (en) | Sample expansion and consistency discrimination method for improving spectral image supervision classification performance | |
CN106909884A (en) | A kind of hand region detection method and device based on hierarchy and deformable part sub-model | |
CN112766145A (en) | Method and device for identifying dynamic facial expressions of artificial neural network | |
CN109146902A (en) | A kind of profile testing method based on color antagonism receptive field and monochrome channel | |
CN105957067A (en) | Color image edge detection method based on color difference | |
CN109766860A (en) | Method for detecting human face based on improved Adaboost algorithm | |
CN107067408B (en) | Image contour detection method for simulating human eye micromotion | |
CN109978898A (en) | Profile testing method based on vector field energy balane |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |