CN107067407B - Contour detection method based on non-classical receptive field and linear nonlinear modulation - Google Patents

Contour detection method based on non-classical receptive field and linear nonlinear modulation Download PDF

Info

Publication number
CN107067407B
CN107067407B CN201710230469.8A CN201710230469A CN107067407B CN 107067407 B CN107067407 B CN 107067407B CN 201710230469 A CN201710230469 A CN 201710230469A CN 107067407 B CN107067407 B CN 107067407B
Authority
CN
China
Prior art keywords
receptive field
classical receptive
central neuron
central
neuron
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.)
Active
Application number
CN201710230469.8A
Other languages
Chinese (zh)
Other versions
CN107067407A (en
Inventor
林川
曹以隽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangxi University of Science and Technology
Original Assignee
Guangxi University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangxi University of Science and Technology filed Critical Guangxi University of Science and Technology
Priority to CN201710230469.8A priority Critical patent/CN107067407B/en
Publication of CN107067407A publication Critical patent/CN107067407A/en
Application granted granted Critical
Publication of CN107067407B publication Critical patent/CN107067407B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Medicines Containing Antibodies Or Antigens For Use As Internal Diagnostic Agents (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention provides a contour detection method based on a non-classical receptive field and linear nonlinear modulation, which comprises the following steps: A. inputting a to-be-detected image subjected to gray level processing; B. carrying out Gabor filtering on an image to be detected to obtain Gabor energy values of all the pixel points in all directions; C. constructing an X cell and Y cell simulation model in retinal ganglion cells; D. calculating to obtain the stimulation response of each central neuron corresponding to the X cell under the non-classical receptive field; E. calculating to obtain the stimulation response of each central neuron corresponding to the Y cell under the non-classical receptive field; F. respectively calculating the stimulus response of the central neuron of X, Y cells modulated by the classical receptive field and the non-classical receptive field, and adding the stimulus response to serve as corresponding contour values; G. and processing the contour value of each pixel point to obtain a final contour value. The method overcomes the defect of low outline recognition rate in the prior art, and has the characteristics of good simulation effect and high outline recognition rate.

Description

Contour detection method based on non-classical receptive field and linear nonlinear modulation
Technical Field
The invention relates to the field of image processing, in particular to contour detection methods based on non-classical receptive field and linear nonlinear modulation.
Background
Contours define the shape of the target, contours are , an important task in target recognition, while target contours obtained from cluttered scenes are important and rather difficult tasks, mainly because there are usually a lot of edges of the texture background around contours, so this work mainly needs to exclude meaningless edges due to texture regions and preserve target contours.
Rodieck in 1965 proposed a mathematical model of concentric circle antagonistic receptive fields, consisting of central mechanisms with strong excitatory action and peripheral mechanisms with weaker inhibitory action but larger area.A Rodieck model, also known as Gaussian model, with differences of two Gaussians, DOG.1966 Enroth-Cugel and Robson observed that cat GCs can be divided into two classes according to their spatio-temporal sum properties of their responses. class of cellular responses generally fit the Rodieck model, i.e. their receptive field excitatory and inhibitory actions can be linearly additive, called X cells, and class of GC is non-linear for which the Rodieck model is not very suitable, called Y cells.
Disclosure of Invention
The invention aims to provide contour detection methods based on non-classical receptive field and linear nonlinear modulation, overcomes the defects of poor simulation effect and low contour recognition rate in the prior art, and has the characteristics of good simulation effect and high contour recognition rate.
The technical scheme of the invention is as follows: the contour detection method based on the non-classical receptive field and the linear nonlinear modulation comprises the following steps:
A. inputting an image to be detected after gray processing, and taking each pixel point of the image to be detected as a central neuron of a non-classical receptive field;
B. presetting a Gabor filter group with a plurality of direction parameters, and respectively carrying out Gabor filtering on each pixel point in an image to be detected according to each direction parameter to obtain Gabor energy values of each pixel point in each direction; for each pixel point, selecting the maximum value in the Gabor energy values in all directions of the pixel point as the stimulation response of the pixel point by the classical receptive field, namely the stimulation response of the central neuron of the non-classical receptive field, and taking the filtering direction corresponding to the maximum value as the optimal angle of the pixel point, namely the optimal angle of the central neuron of the non-classical receptive field;
C. constructing an X cell and Y cell simulation model in retinal ganglion cells;
D. response calculation for X cell pathway:
based on each non-classical receptive field central neuron, calculating a spatial sum modulation weight of each non-classical receptive field to the central neuron, classifying the spatial sum modulation weight into , calculating a distance weight response of each non-classical receptive field to the central neuron, and multiplying the distance weight response of each non-classical receptive field to the central neuron by a corresponding classified spatial sum modulation weight to obtain a stimulation response of each central neuron corresponding to the X cell by the non-classical receptive field;
E. response calculation for Y cell pathway:
uniformly arranging a plurality of directions along the outer edge of a central neuron of a non-classical receptive field at intervals, presetting a plurality of base points in each preset direction, calculating the space sum modulation weight of the base points of each non-classical receptive field to the central neuron, classifying the space sum modulation weight into , calculating the distance weight response of the base points of each non-classical receptive field to the central neuron, and multiplying the distance weight response of the base points of each non-classical receptive field to the central neuron by the corresponding space sum modulation weight classified into to obtain the stimulus response of each central neuron corresponding to a Y cell to the non-classical receptive field;
F. respectively solving the stimulus responses of the central neurons corresponding to the X cells and the Y cells, which are jointly modulated by the classical receptive field and the non-classical receptive field, correcting the stimulus responses of the joint modulation corresponding to the X cells and the Y cells, and adding the corrected stimulus responses, wherein the added value is used as the contour value of the pixel point corresponding to the central neuron;
G. and carrying out non-maximum suppression and double-threshold processing on the contour value of each pixel point to obtain the final contour value of each pixel point.
Preferably, in step D:
respectively calculating to obtain a contrast weight function of each non-classical receptive field to the central neuron, respectively calculating to obtain an orientation weight function of each non-classical receptive field central neuron according to the optimal angle corresponding to each non-classical receptive field central neuron, further calculating a space sum modulation weight of each non-classical receptive field to the central neuron, and classifying the space sum modulation weight into ;
and calculating the distance weight function of each non-classical receptive field to the central neuron, and convolving the distance weight function with the stimulus response of the central neuron subjected to the classical receptive field to obtain the distance weight response of each non-classical receptive field to the central neuron.
Preferably, in step E:
respectively calculating to obtain a contrast ratio weight function of each base point to the central nerve cell of the base point, respectively calculating to obtain a direction weight function of the central nerve cell of each non-classical receptive field according to the angle of each base point, further calculating to obtain a space sum modulation weight of each base point of the non-classical receptive field to the central nerve cell of the base point, and classifying the space sum modulation weight into ;
calculating a distance weight function of the base point of each non-classical receptive field to the central neuron of the non-classical receptive field, and multiplying the distance weight function by the stimulus response of the central neuron under the classical receptive field to obtain the distance weight response of the base point of each non-classical receptive field to the central neuron of the non-classical receptive field;
and multiplying the distance weight response of the base point of each non-classical receptive field to the central neuron by the corresponding spatial summation modulation weight classified as to obtain the stimulation response of each central neuron corresponding to the Y cell by the non-classical receptive field.
Preferably, in the step F, the stimulation responses of the central neurons corresponding to the X cell and the Y cell modulated by the non-classical receptive field are multiplied by the inhibition coefficient, and then the product is subtracted from the stimulation responses of the central neurons modulated by the classical receptive field to obtain the jointly modulated stimulation responses corresponding to the X cell and the Y cell.
Preferably, the step B is as follows:
the two-dimensional Gabor function expression of the Gabor filter bank is as follows:
Figure GDA0002154010320000031
wherein
Figure GDA0002154010320000032
Gamma is constants representing the ratio of the long axis to the short axis of the elliptical field, the parameter lambda is the wavelength, sigma is the bandwidth of the central region of the DoG template, 1/lambda is the spatial frequency of the cosine function,
Figure GDA0002154010320000033
is a phase angle parameter, theta is an angle parameter of the Gabor filtering;
i (x, y) is an image to be detected and is a convolution operator;
the Gabor energy value was calculated as follows:
Figure GDA0002154010320000035
Figure GDA0002154010320000036
wherein theta isiFor a certain angle, N, of Gabor filteringθThe number of the angles of the Gabor filtering is shown;
Figure GDA0002154010320000037
Figure GDA0002154010320000038
ec (x, y) is the maximum value of the Gabor filtering energy value of each angle of the pixel point (x, y),
Figure GDA0002154010320000039
and the filter angle corresponding to Ec (x, y) is taken as the optimal angle of the pixel point (x, y).
Preferably, the step D is specifically as follows:
for X cells:
the expression of the spatial summation modulation weight of the non-classical receptive field to the central neuron is as follows:
inhgox(x,y)＝∑x′∑y′wgx(x,y)wox(x+x′,y+y′)(7)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgox(x, y) is the spatial sum modulation weight of the non-classical receptive field to the central neuron, wgx(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, wox(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgxthe expression (x, y) is:
Figure GDA0002154010320000041
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woxthe expression of (x + x ', y + y') is:
Figure GDA0002154010320000042
where ω is max (ω)1，ω2)；
Figure GDA0002154010320000043
Figure GDA0002154010320000044
Wherein
Figure GDA0002154010320000045
The optimal angle for the central neuron A (x, y),the optimal angle of any neurons B (x + x ', y + y') except the central neuron in the non-classical receptive field, namely the filtering angle corresponding to the maximum gabor energy value of neuron A, B,
Figure GDA0002154010320000047
is the deflection angle of the connecting line of the central neuron A and the neuron B;
Figure GDA0002154010320000048
modulating the weights for the normalized space sum of ;
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
inhdx(x,y)＝Ec(x,y)·wd(x,y) (10)；
wherein inhdx(x, y) is the distance weighted response of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure GDA0002154010320000051
preferably, the step E is specifically as follows:
for Y cells:
the coordinate of each base point is (Δ x)i,j,Δyi,j)，Δxi,j＝dicosφj、Δyi,j＝disinφj；
Wherein d isi＝1,2...5＝(2σ,4σ,6σ,8σ,10σ)，φj＝1,2,...8＝(0,π/4,π/2,3π/4,π,5/4,3π/2,7π/4)；
The energy and angle of each base point are calculated as follows:
Figure GDA0002154010320000052
wherein G isσ(x′,y′；σG) Is a Gaussian function, σG＝σ，x′≤3σG，y′≤3σG(ii) a T is a multiplier of the scale parameter;
the expression of the space summation modulation weight of each base point to the central neuron in the non-classical receptive field is as follows:
inhgoy(x,y)＝∑x′∑y′wgy(x,y)woy(x+x′,y+y′)(14)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgoy(x, y) is the space sum modulation weight of each base point in the non-classical receptive field to the central neuron, wgy(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, woy(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgythe expression (x, y) is:
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woythe expression of (x, y) is:
Figure GDA0002154010320000055
wherein ω is3＝max(ω4，ω5)；
Figure GDA0002154010320000061
WhereinThe optimal angle for the central neuron A (x, y),
Figure GDA0002154010320000064
is the angle of the base point B of any except the central neuron in the non-classical receptive field,
Figure GDA0002154010320000065
is the deviation angle of the connecting line of the central neuron A and the base point neuron B;
modulating the weights for the normalized space sum of ;
Figure GDA0002154010320000067
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
Figure GDA0002154010320000068
wherein inhdy(x, y) is the distance weighted response of the base point of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
wherein, wdThe expression of (x, y) is:
wherein the content of the first and second substances,
Figure GDA00021540103200000610
Figure GDA00021540103200000611
wherein | · | purple sweet1Is (L)1) Norm, H (DoG (x, y)) is a function taking a positive value, and DoG (x, y) is an expression corresponding to the DoG template;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure GDA00021540103200000612
preferably, the step F is specifically as follows:
for X cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Rx(x,y)＝H(Ec(x,y)-αinhx(x,y)) (21)；
wherein R isx(X, y) is the stimulation response of the central neuron corresponding to the X cell modulated by the combination of the classical receptive field and the non-classical receptive field, and α is the inhibition coefficient corresponding to the X cell;
for Y cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Ry(x,y)＝H(Ec(x,y)-βinhy(x,y)) (22)；
wherein R isy(x, Y) is the stimulation response of the central neuron corresponding to the Y cell modulated by the combination of the classical receptive field and the non-classical receptive field, and β is the inhibition coefficient corresponding to the Y cell;
the positive function is
Figure GDA0002154010320000071
The contour value expression of the pixel points corresponding to the central neuron is as follows:
R(x,y)＝Rx(x,y)+Ry(x,y) (23)；
wherein, R (x, y) is the contour value of the pixel point corresponding to the central neuron.
The method simulates the sensing response of the X cell, and simulates the sensing response of the Y cell by using a multiplier of a scale parameter based on the characteristic that the sensing field of the Y cell is larger than that of the X cell; the method has the advantages that the base points in multiple directions are set innovatively, Gaussian weighting integration is carried out on the small circular area of the base point, so that simulation of Y cell sensory response is realized, the set preprocessing of the base points can better store stable contour information; the region information around the base point is integrated through Gaussian weighting, so that simulation of the Y cell sensing response property can be realized, high-scale calculation amount can be effectively reduced, and the calculation efficiency is improved.
Drawings
FIG. 1 is a flow chart of a contour detection method of the present invention
FIG. 2 is a comparison graph of the detection effects of the method of embodiment 1 and the profile detection model of document 1
FIG. 3 is a comparison of the detection parameters of the profile detection model of the document 1 and the method of example 1
FIG. 4 is a schematic diagram of a base point of the contour detection method of the present invention
Detailed Description
Example 1
The contour detection method based on the non-classical receptive field and the linear nonlinear modulation provided by the embodiment comprises the following steps:
A. inputting an image to be detected after gray processing, and taking each pixel point of the image to be detected as a central neuron of a non-classical receptive field;
B. presetting a Gabor filter group with a plurality of direction parameters, and respectively carrying out Gabor filtering on each pixel point in an image to be detected according to each direction parameter to obtain Gabor energy values of each pixel point in each direction; for each pixel point, selecting the maximum value in the Gabor energy values in all directions of the pixel point as the stimulation response of the pixel point by the classical receptive field, namely the stimulation response of the central neuron of the non-classical receptive field, and taking the filtering direction corresponding to the maximum value as the optimal angle of the pixel point, namely the optimal angle of the central neuron of the non-classical receptive field;
the step B is specifically as follows:
the two-dimensional Gabor function expression of the Gabor filter bank is as follows:
Figure GDA0002154010320000081
wherein
Figure GDA0002154010320000082
Gamma is constants representing the ratio of the long axis to the short axis of the elliptical field, the parameter lambda is the wavelength, sigma is the bandwidth of the central region of the DoG template, 1/lambda is the spatial frequency of the cosine function,
Figure GDA0002154010320000083
is a phase angle parameter, theta is an angle parameter of the Gabor filtering;
i (x, y) is an image to be detected and is a convolution operator;
the Gabor energy value was calculated as follows:
Figure GDA0002154010320000085
Figure GDA0002154010320000086
wherein theta isiFor a certain angle, N, of Gabor filteringθThe number of the angles of the Gabor filtering is shown;
Figure GDA0002154010320000087
Figure GDA0002154010320000088
ec (x, y) is the maximum value of the Gabor filtering energy value of each angle of the pixel point (x, y),
Figure GDA0002154010320000089
the filtering angle corresponding to Ec (x, y) is taken as the optimal angle of the pixel point (x, y);
C. constructing an X cell and Y cell simulation model in retinal ganglion cells;
D. response calculation for X cell pathway:
respectively calculating a contrast weight function of each non-classical receptive field to the central neuron based on each non-classical receptive field central neuron, respectively calculating an orientation weight function of each non-classical receptive field central neuron according to an optimal angle corresponding to each non-classical receptive field central neuron, further calculating a space sum modulation weight of each non-classical receptive field to the central neuron, and classifying the space sum modulation weight, simultaneously calculating a distance weight function of each non-classical receptive field to the central neuron, multiplying the distance weight function and a stimulation response of the central neuron under the classical receptive field to the central neuron to obtain a distance weight response of each non-receptive field to the central neuron, and then convolving the distance weight response of each non-classical receptive field to the central neuron and the corresponding classified space sum modulation weight to obtain a stimulation response of each central neuron under the non-classical receptive field corresponding to the X cell;
for X cells:
the expression of the spatial summation modulation weight of the non-classical receptive field to the central neuron is as follows:
inhgox(x,y)＝∑x′∑y′wgx(x,y)wox(x+x′,y+y′) (7)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgox(x, y) is the spatial sum modulation weight of the non-classical receptive field to the central neuron, wgx(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, wox(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgxthe expression (x, y) is:
Figure GDA0002154010320000091
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woxthe expression of (x + x ', y + y') is:
Figure GDA0002154010320000092
where ω is max (ω)1，ω2)；
Figure GDA0002154010320000093
Figure GDA0002154010320000094
Wherein
Figure GDA0002154010320000095
The optimal angle for the central neuron A (x, y),
Figure GDA0002154010320000096
the optimal angle of any neurons B (x + x ', y + y') except the central neuron in the non-classical receptive field, namely the filtering angle corresponding to the maximum gabor energy value of neuron A, B,
Figure GDA0002154010320000097
is the deflection angle of the connecting line of the central neuron A and the neuron B;
Figure GDA0002154010320000098
modulating the weights for the normalized space sum of ;
Figure GDA0002154010320000101
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
inhdx(x,y)＝Ec(x,y)·wd(x,y)(10)；
wherein inhdx(x, y) is the distance weighted response of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure GDA0002154010320000102
E. response calculation for Y cell pathway:
respectively calculating to obtain a contrast ratio weight function of each base point to the central neurons of the non-classical receptive field, respectively calculating to obtain a direction weight function of each central neuron of the non-classical receptive field according to the angle of each base point, further calculating to obtain a space sum modulation weight of each base point of the non-classical receptive field to the central neurons, and grouping the space sum modulation weights into ;
calculating a distance weight function of the base point of each non-classical receptive field to the central neuron of the non-classical receptive field, and multiplying the distance weight function by the stimulus response of the central neuron under the classical receptive field to obtain the distance weight response of the base point of each non-classical receptive field to the central neuron of the non-classical receptive field;
convolving the distance weight response of the central neuron with the corresponding space sum modulation weight classified as through the base point of each non-classical receptive field to obtain the stimulation response of each central neuron corresponding to the Y cell under the non-classical receptive field;
the coordinate of each base point is (Δ x)i,j,Δyi,j)，Δxi,j＝dicosφj、Δyi,j＝disinφj；
Wherein d isi＝1,2...5＝(2σ,4σ,6σ,8σ,10σ)，φj＝1,2,...8＝(0,π/4,π/2,3π/4,π,5/4,3π/2,7π/4)；
The energy and angle of each base point are calculated as follows:
Figure GDA0002154010320000103
Figure GDA0002154010320000104
wherein G isσ(x′,y′；σG) Is a Gaussian function, σG＝σ，x′≤3σG，y′≤3σG(ii) a T is a multiplier of the scale parameter;
the expression of the space summation modulation weight of each base point to the central neuron in the non-classical receptive field is as follows:
inhgoy(x,y)＝∑x′∑y′wgy(x,y)woy(x+x′,y+y′)(14)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgoy(x, y) is the space sum modulation weight of each base point in the non-classical receptive field to the central neuron, wgy(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, woy(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgythe expression (x, y) is:
Figure GDA0002154010320000111
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woythe expression of (x, y) is:
Figure GDA0002154010320000112
wherein ω is3＝max(ω4，ω5)；
Figure GDA0002154010320000113
Figure GDA0002154010320000114
Wherein
Figure GDA0002154010320000115
The optimal angle for the central neuron A (x, y),
Figure GDA0002154010320000116
is the angle of the base point B of any except the central neuron in the non-classical receptive field,
Figure GDA0002154010320000117
is the deviation angle of the connecting line of the central neuron A and the base point neuron B;
Figure GDA0002154010320000118
modulating the weights for the normalized space sum of ;
Figure GDA0002154010320000119
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
Figure GDA00021540103200001110
wherein inhdy(x, y) is the distance weighted response of the base point of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
wherein, wdThe expression of (x, y) is:
Figure GDA0002154010320000121
wherein the content of the first and second substances,
Figure GDA0002154010320000122
Figure GDA0002154010320000123
wherein | · | purple sweet1Is (L)1) Norm, H (DoG (x, y)) is a function taking a positive value, and DoG (x, y) is an expression corresponding to the DoG template;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure GDA0002154010320000124
F. respectively multiplying the stimulation response of each central neuron corresponding to the X cell and the Y cell modulated by the non-classical receptive field by an inhibition coefficient, subtracting the product from the stimulation response of the central neuron modulated by the classical receptive field to respectively obtain the stimulation response of the combined modulation corresponding to the X cell and the Y cell, taking the positive values of the stimulation response of the combined modulation corresponding to the X cell and the Y cell, adding the positive values, and taking the added value as the contour value of the pixel point corresponding to the central neuron;
for X cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Rx(x,y)＝H(Ec(x,y)-αinhx(x,y)) (21)；
wherein R isx(X, y) is the stimulation response of the central neuron corresponding to the X cell modulated by the combination of the classical receptive field and the non-classical receptive field, and α is the inhibition coefficient corresponding to the X cell;
for Y cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Ry(x,y)＝H(Ec(x,y)-βinhy(x,y)) (22)；
wherein R isy(X, Y) is the stimulation response of the central neuron corresponding to the Y cell modulated by the combination of the classical receptive field and the non-classical receptive field, and β is the inhibition coefficient corresponding to the X cell;
the positive function is
Figure GDA0002154010320000125
The contour value expression of the pixel points corresponding to the central neuron is as follows:
R(x,y)＝Rx(x,y)+Ry(x,y) (23)；
wherein, R (x, y) is the contour value of the pixel point corresponding to the central neuron.
The effectiveness of the contour detection method of this embodiment is compared with the effectiveness of the contour detection isotropic model and anisotropic model provided in document 1, wherein the isotropic model and anisotropic model in document 1 are selected for effectiveness comparison, and document 1 is as follows:
document 1: grigoresecu C, Petkov N, Westenberg M. content detection based on systematic iterative field inhibition [ J ]. IEEE Transactions on Imageprocessing,2003,12(7): 729-;
to ensure the effectiveness of the comparison, the following contour integration is performed by the same non-maximum suppression method as that in document 1, including two thresholds th,tlIs set to tl＝0.5thCalculated from a threshold quantile p;
wherein the performance evaluation index P adopts the following criteria given in document 1:
Figure GDA0002154010320000131
in the formula nTP、nFP、nFNRespectively representing the number of detected correct contours, error contours and missing contours, and the evaluation index P is set to be [0,1 ]]The closer to 1, the better the contour detection, and in addition, the tolerance is defined as: all detected within 5 x 5 neighbourhoods are counted as correct detections.
Selecting 3 pairs of classical images of rhinoceros, lions and elephants for effectiveness comparison, and respectively adopting an isotropic model and an anisotropic model in document 1 and an embodiment 1 method to carry out contour detection on the 3 images, wherein the parameter set selected by the embodiment 1 method is shown in table 1,
table 1 example 1 parameter set table
Figure GDA0002154010320000132
The isotropic model and anisotropic model in document 1 use 80 sets of parameters, α ═ {1.0, 1.2}, σ ═ {1.4, 1.6, 1.8, 2.0, 2.2, 2.4, 2.6, 2.8}, p ═ 0.5, 0.4, 0.3, 0.2, 0.1 };
selecting sets of parameters with best effect in the method of embodiment 1 for comparison, wherein a contour extraction effect graph is shown in figure 2, and a corresponding performance index graph is shown in figure 3, as can be seen from figures 2 and 3, the method of embodiment 1 is superior to the isotropic model and the anisotropic model in the document 1 in terms of the contour extraction effect and the performance index parameters, figure 4 is a schematic diagram of a Y cell receptive field base point of embodiment 1, and a black point is the base point;
wherein, the table 2 is a partial parameter table corresponding to the result chart of the embodiment 1, and the rest parameters refer to the data in the table 1; tables 3 and 4 are parameter tables corresponding to the results of the isotropic model and the anisotropic model, respectively.
Table 2 partial parameter table corresponding to result chart of example 1
Figure GDA0002154010320000141
TABLE 3 parameter tables corresponding to results of the isotropic model
Figure GDA0002154010320000142
TABLE 4 parameter table corresponding to result graph of anisotropy model
Figure GDA0002154010320000143

Claims (8)

1. The contour detection method based on the non-classical receptive field and the linear nonlinear modulation is characterized by comprising the following steps:
A. inputting an image to be detected after gray processing, and taking each pixel point of the image to be detected as a central neuron of a non-classical receptive field;
B. presetting a Gabor filter group with a plurality of direction parameters, and respectively carrying out Gabor filtering on each pixel point in an image to be detected according to each direction parameter to obtain Gabor energy values of each pixel point in each direction; for each pixel point, selecting the maximum value in the Gabor energy values in all directions of the pixel point as the stimulation response of the pixel point by the classical receptive field, namely the stimulation response of the central neuron of the non-classical receptive field, and taking the filtering direction corresponding to the maximum value as the optimal angle of the pixel point, namely the optimal angle of the central neuron of the non-classical receptive field;
C. constructing an X cell and Y cell simulation model in retinal ganglion cells;
D. response calculation for X cell pathway:
based on each non-classical receptive field central neuron, calculating a spatial sum modulation weight of each non-classical receptive field to the central neuron, classifying the spatial sum modulation weight into , calculating a distance weight response of each non-classical receptive field to the central neuron, and multiplying the distance weight response of each non-classical receptive field to the central neuron by a corresponding classified spatial sum modulation weight to obtain a stimulation response of each central neuron corresponding to the X cell by the non-classical receptive field;
E. response calculation for Y cell pathway:
uniformly arranging a plurality of directions along the outer edge of a central neuron of a non-classical receptive field at intervals, presetting a plurality of base points in each preset direction, calculating the space sum modulation weight of the base points of each non-classical receptive field to the central neuron, classifying the space sum modulation weight into , calculating the distance weight response of the base points of each non-classical receptive field to the central neuron, and multiplying the distance weight response of the base points of each non-classical receptive field to the central neuron by the corresponding space sum modulation weight classified into to obtain the stimulus response of each central neuron corresponding to a Y cell to the non-classical receptive field;
F. respectively solving the stimulus responses of the central neurons corresponding to the X cells and the Y cells, which are jointly modulated by the classical receptive field and the non-classical receptive field, correcting the stimulus responses of the joint modulation corresponding to the X cells and the Y cells, and adding the corrected stimulus responses, wherein the added value is used as the contour value of the pixel point corresponding to the central neuron;
G. and carrying out non-maximum suppression and double-threshold processing on the contour value of each pixel point to obtain the final contour value of each pixel point.
2. The contour detection method based on non-classical receptive field and linear non-linear modulation according to claim 1, characterized by:
in the step D:
respectively calculating to obtain a contrast weight function of each non-classical receptive field to the central neuron, respectively calculating to obtain an orientation weight function of each non-classical receptive field central neuron according to the optimal angle corresponding to each non-classical receptive field central neuron, further calculating a space sum modulation weight of each non-classical receptive field to the central neuron, and classifying the space sum modulation weight into ;
and calculating the distance weight function of each non-classical receptive field to the central neuron, and convolving the distance weight function with the stimulus response of the central neuron subjected to the classical receptive field to obtain the distance weight response of each non-classical receptive field to the central neuron.
3. The method of claim 2, wherein the contour detection based on the non-classical receptive field and the linear non-linear modulation is characterized by:
in the step E:
respectively calculating to obtain a contrast ratio weight function of each base point to the central nerve cell of the base point, respectively calculating to obtain a direction weight function of the central nerve cell of each non-classical receptive field according to the angle of each base point, further calculating to obtain a space sum modulation weight of each base point of the non-classical receptive field to the central nerve cell of the base point, and classifying the space sum modulation weight into ;
calculating a distance weight function of the base point of each non-classical receptive field to the central neuron, and convolving the distance weight function with the stimulus response of the central neuron subjected to the classical receptive field to obtain the distance weight response of the base point of each non-classical receptive field to the central neuron;
and (3) convolving the distance weight response of the central neuron with the corresponding space sum modulation weight classified as through the base point of each non-classical receptive field to obtain the stimulation response of each central neuron corresponding to the Y cell under the non-classical receptive field.
4. The method of claim 3, wherein the contour detection based on the non-classical receptive field and the linear non-linear modulation is characterized by:
and step F, multiplying the stimulus response of each central neuron corresponding to the X cell and the Y cell modulated by the non-classical receptive field by the inhibition coefficient to obtain a product, and subtracting the product from the stimulus response of the central neuron modulated by the classical receptive field to obtain the stimulus response of the X cell and the combined modulation corresponding to the Y cell.
5. The method of claim 4, wherein the contour detection based on the non-classical receptive field and the linear non-linear modulation is as follows:
the step B is specifically as follows:
the two-dimensional Gabor function expression of the Gabor filter bank is as follows:
Figure FDA0002154010310000021
whereinGamma is constants representing the ratio of the long axis to the short axis of the elliptical field, the parameter lambda is the wavelength, sigma is the bandwidth of the central region of the DoG template, 1/lambda is the spatial frequency of the cosine function,
Figure FDA0002154010310000023
is a phase angle parameter, theta is an angle parameter of the Gabor filtering;
Figure FDA0002154010310000024
i (x, y) is an image to be detected and is a convolution operator;
the Gabor energy value was calculated as follows:
Figure FDA0002154010310000031
Figure FDA0002154010310000032
wherein theta isiFor a certain angle, N, of Gabor filteringθThe number of the angles of the Gabor filtering is shown;
Figure FDA0002154010310000033
Figure FDA0002154010310000034
ec (x, y) is the maximum value of the Gabor filtering energy value of each angle of the pixel point (x, y),
Figure FDA0002154010310000035
and the filter angle corresponding to Ec (x, y) is taken as the optimal angle of the pixel point (x, y).
6. The method of claim 5, wherein the contour detection based on non-classical receptive field and linear non-linear modulation is characterized by:
the step D is specifically as follows:
for X cells:
the expression of the spatial summation modulation weight of the non-classical receptive field to the central neuron is as follows:
inhgox(x,y)＝∑x′∑y′wgx(x,y)wox(x+x′,y+y′) (7)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgox(x, y) is the spatial sum modulation weight of the non-classical receptive field to the central neuron, wgx(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, wox(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgxthe expression (x, y) is:
Figure FDA0002154010310000036
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woxthe expression of (x + x ', y + y') is:
where ω is max (ω)1，ω2)；
Figure FDA0002154010310000043
Wherein
Figure FDA0002154010310000044
The optimal angle for the central neuron A (x, y),
Figure FDA0002154010310000045
the optimal angle of any neurons B (x + x ', y + y') except the central neuron in the non-classical receptive field, namely the filtering angle corresponding to the maximum gabor energy value of neuron A, B,
Figure FDA0002154010310000046
is the deflection angle of the connecting line of the central neuron A and the neuron B;
Figure FDA0002154010310000047
modulating the weights for the normalized space sum of ;
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
inhdx(x,y)＝Ec(x,y)·wd(x,y) (10)；
wherein inhdx(x, y) is the distance weighted response of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure FDA0002154010310000049
7. the method of claim 6, wherein the contour detection based on non-classical receptive field and linear non-linear modulation is characterized by:
the step E is specifically as follows:
for Y cells:
the coordinate of each base point is (Δ x)i,j,Δyi,j)，Δxi,j＝dicosφj、Δyi,j＝disinφj；
Wherein d isi＝1,2...5＝(2σ,4σ,6σ,8σ,10σ)，φj＝1,2,...8＝(0,π/4,π/2,3π/4,π,5/4,3π/2,7π/4)；
The energy and angle of each base point are calculated as follows:
Figure FDA00021540103100000410
wherein G isσ(x′,y′；σG) Is a Gaussian function, σG＝σ，x′≤3σG，y′≤3σG(ii) a T is a multiplier of the scale parameter;
the expression of the space summation modulation weight of each base point to the central neuron in the non-classical receptive field is as follows:
inhgoy(x,y)＝∑x′∑y′wgy(x,y)woy(x+x′,y+y′) (14)；
wherein-3 k σ < x' <3k σ; -3k σ < y' <3k σ;
wherein inhgoy(x, y) is the space sum modulation weight of each base point in the non-classical receptive field to the central neuron, wgy(x, y) is the contrast weighting function of the non-classical receptive field on the central neuron, woy(x + x ', y + y') is the orientation weight function of the non-canonical receptive field to the central neuron;
wgythe expression (x, y) is:
wherein E isAVGThe average value of the response of each pixel point of the image to be detected by the classical receptive field stimulus is the average value of the Ec (x, y) value of each pixel point of the image to be detected;
woythe expression of (x, y) is:
Figure FDA0002154010310000052
wherein ω is3＝max(ω4，ω5)；
Figure FDA0002154010310000053
Figure FDA0002154010310000054
Wherein
Figure FDA0002154010310000055
The optimal angle for the central neuron A (x, y),is the angle of the base point B of any except the central neuron in the non-classical receptive field,
Figure FDA0002154010310000057
is the deviation angle of the connecting line of the central neuron A and the base point neuron B;
modulating the weights for the normalized space sum of ;
Figure FDA0002154010310000059
the distance weight response of the non-classical receptive field to the central neuron is expressed as follows:
Figure FDA00021540103100000510
wherein inhdy(x, y) is the distance weighted response of the base point of the non-classical receptive field to the central neuron, wd(x, y) is a distance weight function of the non-canonical receptive field to the central neuron;
wherein, wdThe expression of (x, y) is:
Figure FDA0002154010310000061
wherein the content of the first and second substances,
Figure FDA0002154010310000062
Figure FDA0002154010310000063
wherein | · | purple sweet1Is (L)1) Norm, H (DoG (x, y)) is a function taking a positive value, and DoG (x, y) is an expression corresponding to the DoG template;
the expression of the stimulation response of the central neuron by the non-classical receptive field is as follows:
Figure FDA0002154010310000064
8. the method of claim 7, wherein the contour detection based on non-classical receptive field and linear non-linear modulation is characterized by:
the step F is specifically as follows:
for X cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Rx(x,y)＝H(Ec(x,y)-αinhx(x,y)) (21)；
wherein R isx(X, y) is the stimulation response of the central neuron corresponding to the X cell modulated by the combination of the classical receptive field and the non-classical receptive field, and α is the inhibition coefficient corresponding to the X cell;
for Y cells:
the expression of the stimulus response of the central neuron modulated by the classical receptive field and the non-classical receptive field is as follows:
Ry(x,y)＝H(Ec(x,y)-βinhy(x,y)) (22)；
wherein R isy(x, Y) is the stimulation response of the central neuron corresponding to the Y cell modulated by the combination of the classical receptive field and the non-classical receptive field, and β is the inhibition coefficient corresponding to the Y cell;
said function taking positive value is
Figure FDA0002154010310000065
The contour value expression of the pixel points corresponding to the central neuron is as follows:
R(x,y)＝Rx(x,y)+Ry(x,y) (23)；
wherein, R (x, y) is the contour value of the pixel point corresponding to the central neuron.
CN201710230469.8A 2017-04-11 2017-04-11 Contour detection method based on non-classical receptive field and linear nonlinear modulation Active CN107067407B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710230469.8A CN107067407B (en) 2017-04-11 2017-04-11 Contour detection method based on non-classical receptive field and linear nonlinear modulation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710230469.8A CN107067407B (en) 2017-04-11 2017-04-11 Contour detection method based on non-classical receptive field and linear nonlinear modulation

Publications (2)

Publication Number Publication Date
CN107067407A CN107067407A (en) 2017-08-18
CN107067407B true CN107067407B (en) 2020-01-31

Family

ID=59601812

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710230469.8A Active CN107067407B (en) 2017-04-11 2017-04-11 Contour detection method based on non-classical receptive field and linear nonlinear modulation

Country Status (1)

Country Link
CN (1) CN107067407B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210493B (en) * 2019-04-30 2021-03-19 中南民族大学 Contour detection method and system based on non-classical receptive field modulation neural network
CN111062957B (en) * 2019-10-28 2024-02-09 广西科技大学鹿山学院 Non-classical receptive field contour detection method
CN111179294B (en) * 2019-12-30 2020-09-15 广西科技大学 Bionic type contour detection method based on X, Y parallel visual channel response
CN111161291A (en) * 2019-12-31 2020-05-15 广西科技大学 Contour detection method based on target depth of field information
CN113192092B (en) * 2021-05-07 2023-04-07 广西科技大学 Contour detection method for simulating fusion of properties of receptor field of XYW cell

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254304A (en) * 2011-06-17 2011-11-23 电子科技大学 Method for detecting contour of target object
CN104484667A (en) * 2014-12-30 2015-04-01 华中科技大学 Contour extraction method based on brightness characteristic and contour integrity
CN106033610A (en) * 2016-03-22 2016-10-19 广西科技大学 Contour detection method based on non-classical receptive field space summation modulation
CN106033607A (en) * 2015-07-24 2016-10-19 广西科技大学 Target contour detection method of biomimetic jumping eye movement information processing mechanism
CN106033606A (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
WO2016195496A2 (en) * 2015-06-05 2016-12-08 Universiteit Van Amsterdam Deep receptive field networks
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

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102111000B1 (en) * 2011-08-25 2020-05-14 코넬 유니버시티 Retinal encoder for machine vision

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254304A (en) * 2011-06-17 2011-11-23 电子科技大学 Method for detecting contour of target object
CN104484667A (en) * 2014-12-30 2015-04-01 华中科技大学 Contour extraction method based on brightness characteristic and contour integrity
WO2016195496A2 (en) * 2015-06-05 2016-12-08 Universiteit Van Amsterdam Deep receptive field networks
CN106033607A (en) * 2015-07-24 2016-10-19 广西科技大学 Target contour detection method of biomimetic jumping eye movement information processing mechanism
CN106033606A (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
CN106033610A (en) * 2016-03-22 2016-10-19 广西科技大学 Contour detection method based on non-classical receptive field space summation modulation
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 (4)

* Cited by examiner, † Cited by third party
Title
A Cooperation and Competition Based Simple Cell Receptive Field Model and Study of Feed-Forward Linear and Nonlinear Contributions to Orientation Selectivity;Basabi Bhaumik 等;《Computational Neuroscience》;20030331;第14卷(第2期);第211-227页 *
Improved contour detection model with spatial summation properties based on nonclassical receptive field;Chuan Lin 等;《J. of Electronic Imaging》;20160802;第25卷(第4期);第1-10页 *
一种仿生物视觉感知的视频轮廓检测方法;谢昭;《自动化学报》;20151031;第41卷(第10期);第1814-1824页 *
基于视通路多感受野朝向性关联的轮廓检测方法;李康群 等;《中国生物医学工程学报》;20170228;第36卷(第1期);第1-11页 *

Also Published As

Publication number Publication date
CN107067407A (en) 2017-08-18

Similar Documents

Publication Publication Date Title
CN107067407B (en) Contour detection method based on non-classical receptive field and linear nonlinear modulation
Kim et al. Median filtered image restoration and anti-forensics using adversarial networks
CN104361328B (en) A kind of facial image normalization method based on adaptive multiple row depth model
CN107767387B (en) Contour detection method based on variable receptive field scale global modulation
CN106033610B (en) Profile testing method based on the modulation of non-classical receptive field spatial summation
CN104537630A (en) Method and device for image beautifying based on age estimation
CN104268593A (en) Multiple-sparse-representation face recognition method for solving small sample size problem
CN103208097B (en) Filtering method is worked in coordination with in the principal component analysis of the multi-direction morphosis grouping of image
CN105095836B (en) A kind of skin-texture detection method and device based on Gabor characteristic
CN103593652B (en) A kind of cucumber disease recognition methods processed based on cucumber leaves symptomatic picture
CN110232390B (en) Method for extracting image features under changed illumination
CN102289670B (en) Image characteristic extraction method with illumination robustness
Premaratne et al. Image matching using moment invariants
Setiawan et al. Lake edge detection using Canny algorithm and Otsu thresholding
CN109949324B (en) Contour detection method based on non-classical receptive field nonlinear subunit response
CN107742302B (en) Contour detection method based on primary visual cortex multi-scale contour fusion
CN104408472B (en) Classification of Polarimetric SAR Image method based on Wishart and SVM
CN104091145A (en) Human palm vein feature image acquisition method
CN109872326B (en) Contour detection method based on deep reinforced network jump connection
Chaki et al. Recognition of whole and deformed plant leaves using statistical shape features and neuro-fuzzy classifier
CN104616280A (en) Image registration method based on maximum stable extreme region and phase coherence
CN105678249A (en) Face identification method aiming at registered face and to-be-identified face image quality difference
CN111179294B (en) Bionic type contour detection method based on X, Y parallel visual channel response
CN107330431B (en) Significance detection optimization method based on K-means cluster fitting
CN105631441A (en) Human face recognition method

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