CN103957395B - There is the color constancy method of adaptive ability - Google Patents
There is the color constancy method of adaptive ability Download PDFInfo
- Publication number
- CN103957395B CN103957395B CN201410190874.8A CN201410190874A CN103957395B CN 103957395 B CN103957395 B CN 103957395B CN 201410190874 A CN201410190874 A CN 201410190874A CN 103957395 B CN103957395 B CN 103957395B
- Authority
- CN
- China
- Prior art keywords
- sigma
- passage
- green
- red
- subprovince
- 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
Links
Landscapes
- Color Image Communication Systems (AREA)
- Facsimile Image Signal Circuits (AREA)
- Processing Of Color Television Signals (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of color constancy method with adaptive ability, especially by non-classical receptive field list antagonism model, the interaction of the inhibitory action of periphery and the removal of inhibit function of subprovince under different coefficient of sensitivity is utilized automatically to adapt to different scene image, by take under non-standard illumination have colour cast color image restoration become standard light according to lower shooting without colour cast coloured image, realize the automatic calibration of colour cast color of image, i.e. color constancy.On the image of the different scene of nearly thousand width, illumination on multiple international color constancy database, test proves, the method for this invention has better effect than classical color constancy algorithm.
Description
Technical field
The invention belongs to technical field of image processing, relate to the constant technology of color image color, is a kind of color constancy method based on retina non-classical receptive field vision mechanism.
Background technology
Different from transducers such as video cameras, although the same surface reflection of object enters the coloured light difference of human eye under different light, visual perception can get identical body surface reflectivity properties from different incident light spectrum, therefore when extraneous illumination variation within limits time, the mankind are relatively constant to the perception of object color, constancy to color-aware in this certain limit, is called as color constancy (ColorConstancy).The object color that the transducers such as video camera collect is determined by the transport property of light source, body surface reflectivity properties and transducer self, the object color that the change of light source will greatly change it and collects.Such as, hot illumination on blank sheet of paper, from people can perceive paper still for white different, in machine vision due to transducer collect be from paper reflection ruddiness, what perceive will be red paper.Therefore, need in image procossing to utilize color constancy method, become by the Postprocessing technique taken under non-standard illumination standard light according to the image of lower shooting.At present, different according to basic thought, the method realizing color constancy can be divided into the method based on light source estimation and the large class of the method two based on color invariance.The method estimated based on light source is by estimating evenly or the light source of hypothesis uniform irradiation on object, and then process is carried out to colour cast image recover without colour cast image, its exemplary process has the method based on bayesian theory, the method based on Color Gamut Mapping, method etc. based on image simple statistics feature, and this is also the main stream approach of color constancy in current machine vision.Although these methods can realize color constancy in various degree, all there is certain defect.Method based on bayesian theory be according to bayesian criterion select from many candidate light source there is maximum probability light source as the colour cast light source estimated, require prior information; Method based on Color Gamut Mapping cannot avoid empty solution; Method based on image simple statistics feature needs based on certain a priori assumption, will lose efficacy when colour cast image does not meet this.Method based on color invariance directly processes the local feature of original colour cast image according to some color invariance is theoretical thus directly obtains without colour cast image, and most and physiological mechanism is closely related.
Although Chinese invention patent 200910167730.X make use of non-classical receptive field model equally, it does not consider center, perimeter region, the subprovince different coefficient of sensitivity under different scene, automatically can not adapt to the different images of different characteristic.
Summary of the invention
The object of the invention is the problems referred to above existed to solve prior art, proposing a kind of color constancy method with adaptive ability.
Technical scheme of the present invention is: a kind of color constancy method with adaptive ability, as shown in Figure 2, comprises the following steps:
Step 1: set receptive field size and determine corresponding model parameter:
Setting receptive field center radius, inhibition zone, periphery radius, subprovince radius;
Center, periphery, subprovince gaussian kernel function are:
Wherein, central Gaussian distributed constant σ
c, periphery Gaussian Distribution Parameters σ
s, subprovince Gaussian Distribution Parameters σ
u, be respectively 1/3rd of its corresponding region radius;
Step 2: red component I is extracted respectively to each pixel of colour cast image
r, green component I
g, blue component I
bwith yellow color component I
y, after level and smooth with each component input Gaussian function of each pixel, open P power after the P rank exponential average of output, obtain corresponding P norm, be designated as R, G, B and Y, specific as follows:
R=(mean((Gauss(I
R))
p))
1/p
G=(mean((Gauss(I
G))
p))
1/p
B=(mean((Gauss(I
B))
p))
1/p
Y=(mean((Gauss(I
Y))
p))
1/p,
Wherein, mean () expression is averaging computing;
Step 3: utilize formula:
Calculate red green antagonism passage non-classical receptive field center coefficient of sensitivity A
r1, green red antagonism passage non-classical receptive field center coefficient of sensitivity A
g1, blue yellow antagonism passage non-classical receptive field center coefficient of sensitivity A
b1;
Step 4: according to the excited rejection ratio K of setting, calculate red green antagonism passage non-classical receptive field periphery coefficient of sensitivity A
r2, subprovince coefficient of sensitivity A
r3, green red antagonism passage non-classical receptive field periphery coefficient of sensitivity A
g2, subprovince coefficient of sensitivity A
g3, blue yellow antagonism passage non-classical receptive field periphery coefficient of sensitivity A
b2, subprovince coefficient of sensitivity A
b3:
A
R2=K×A
R1/5,A
R3=A
R2/3
A
G2=K×A
G1/5,A
G3=A
G2/3
A
B2=K×A
B1/5,A
B3=A
B2/3
According to the determined center of step 1, periphery, subprovince gaussian kernel function, according to order from left to right, from top to bottom, each pixel (x, y) of colour cast image is carried out the operation of following step 5 to step 7 successively as the center of a receptive field:
Step 5: according to formula:
Calculate after disinthibiting in red green passage subprovince and respond R
r3(x, y; σ
u), green red passage subprovince responds R after disinthibiting
g3(x, y; σ
u), blue yellow passage subprovince responds R after disinthibiting
b3(x, y; σ
u), wherein, (p, q), for dropping on the point outside Unit Nei Chu center, subprovince, MAX represents and gets higher value in both;
Step 6: according to formula:
Calculate red green passage periphery and suppress R
r2(x, y; σ
s), green red passage periphery suppresses R
g2(x, y; σ
s), blue yellow passage periphery suppresses R
b2(x, y; σ
s), wherein, (p, q) is for dropping on the point in the Surround of periphery;
Step 7: according to formula:
Calculate after district of Hong Lv channel center suppresses and respond R
r1(x, y; σ
c), district of Lv Hong channel center responds R after suppressing
g1(x, y; σ
c), district of Lan Huang channel center responds R after suppressing
b1(x, y; σ
c), wherein, (p, q) is for dropping on the point in the Center of center; Get R
r1(x, y; σ
c), R
g1(x, y; σ
c), R
b1(x, y; σ
c) as the new red, green, blue component of central pixel point (x, y);
Step 8: after Mobility Center pixel (x, y) travels through full figure, to the new red component I of all pixels of image
r, green component I
g, blue component I
b, yellow color component I
ysue for peace respectively, excited rejection ratio K adds 1, and difference iteration on red green, green red, blue yellow antagonism passage, repeats step 4 to 8;
The condition of above-mentioned iteration termination is: often take turns after iteration, subchannel check this passage corresponding color component and, if after when its first derivative and second dervative are all less than the value preset, all passages stop iteration, export as red component I with red green passage
r, green red passage exports as green component I
g, blue yellow passage exports as blue component I
bthe inclined coloured image of synthesizing colourless.
In step 8, its first derivative and second dervative are all less than the value preset, and are specially when first derivative and second dervative all level off to zero, this passage chromatic adaptation curve are described steadily, and this passage stops iteration.
Beneficial effect of the present invention: method of the present invention is by non-classical receptive field list antagonism model, the interaction of the inhibitory action of periphery and the removal of inhibit function of subprovince under different coefficient of sensitivity is utilized automatically to adapt to different scene image, by take under non-standard illumination have colour cast color image restoration become standard light according to lower shooting without colour cast coloured image, realize the automatic calibration of colour cast color of image, i.e. color constancy.On the image of the different scene of nearly thousand width, illumination on multiple international color constancy database, test proves, the method for this invention has better effect than classical color constancy algorithm.
Accompanying drawing explanation
Fig. 1 is the retina non-classical receptive field model being with subprovince of disinthibiting.
Fig. 2 is the schematic flow sheet of the inventive method.
Embodiment
In the vision system of the present embodiment, single Visual Neuron reacts to the stimulation in certain specific region in the visual field, and this region is called as neuronic receptive field.Outside this region, there is the non-classical receptive field of a wider unit's response that affects the nerves.Amphiblestroid receptive field is the structure of periphery, center, has one on a large scale, the district of disinthibiting be made up of multiple subprovince in its periphery, suppresses periphery to the inhibitory action at center, i.e. non-classical receptive field (as shown in Figure 1).To the perception of color from corresponding to L, M, S tri-of red, green, blue three primary colors class cone cell, gangliocyte is passed to through Beale's ganglion cells by after horizontal cell negative feedback ashing, processed respectively by red green, blue yellow antagonism passage, remove the impact of filling the air illumination to a certain extent by the removal of inhibit function of non-classical receptive field, thus realize color constancy.
Embodiment one: test on Gehler storehouse
Comprise 568 pictures international color constancy image data base Gehler storehouse (
http:// www.cs.sfu.ca/ ~ colour/data/shi_gehler/) on testing evaluation is carried out to this method.Receptive field center radius is set to 1, inhibition zone, periphery radius is set to 3, subprovince radius is set to 1, P norm gets 10, test result is as shown in table 1, than classical color constancy algorithm have better effect (angular error represent after process without colour cast image with truly without the gap of colour cast image, less account for color recovery better).
Table 1.Gehler storehouse image algorithms of different angular error median
Embodiment two: test on SFU storehouse
Comprise 31 scene 321 pictures international color constancy image data base SFU storehouse (
http:// www.cs.sfu.ca/ ~ colour/data/colour_constancy_test_images/index.html/) on testing evaluation is carried out to this method.Receptive field center radius is set to 1, inhibition zone, periphery radius is set to 3, subprovince radius is set to 1, norm gets 15, test result is as shown in table 2 below, than classical color constancy algorithm have better effect (angular error represent after process without colour cast image with truly without the gap of colour cast image, less account for color recovery better).
Table 2.SFU storehouse image algorithms of different angular error median
Claims (1)
1. there is a color constancy method for adaptive ability, comprise the following steps:
Step 1: set receptive field size and determine corresponding model parameter:
Setting receptive field center radius, inhibition zone, periphery radius, subprovince radius;
Center, periphery, subprovince gaussian kernel function are:
Wherein, central Gaussian distributed constant σ
c, periphery Gaussian Distribution Parameters σ
s, subprovince Gaussian Distribution Parameters σ
u, be respectively 1/3rd of its corresponding region radius;
Step 2: red component I is extracted respectively to each pixel of colour cast image
r, green component I
g, blue component I
bwith yellow color component I
y, after level and smooth with each component input Gaussian function of each pixel, open P power after the P rank exponential average of output, obtain corresponding P norm, be designated as R, G, B and Y,
R=(mean((Gauss(I
R))
p))
1/p
G=(mean((Gauss(I
G))
p))
1/p
B=(mean((Gauss(I
B))
p))
1/p
Y=(mean((Gauss(I
Y))
p))
1/p,
Wherein, mean () expression is averaging computing;
Step 3: utilize formula:
Calculate red green antagonism passage non-classical receptive field center coefficient of sensitivity A
r1, green red antagonism passage non-classical receptive field center coefficient of sensitivity A
g1, blue yellow antagonism passage non-classical receptive field center coefficient of sensitivity A
b1;
Step 4: according to the excited rejection ratio K of setting, calculate red green antagonism passage non-classical receptive field periphery coefficient of sensitivity A
r2, subprovince coefficient of sensitivity A
r3, green red antagonism passage non-classical receptive field periphery coefficient of sensitivity A
g2, subprovince coefficient of sensitivity A
g3, blue yellow antagonism passage non-classical receptive field periphery coefficient of sensitivity A
b2, subprovince coefficient of sensitivity A
b3:
A
R2=K×A
R1/5,A
R3=A
R2/3
A
G2=K×A
G1/5,A
G3=A
G2/3
A
B2=K×A
B1/5,A
B3=A
B2/3
According to the determined center of step 1, periphery, subprovince gaussian kernel function, according to order from left to right, from top to bottom, each pixel (x, y) of colour cast image is carried out the operation of following step 5 to step 7 successively as the center of a receptive field:
Step 5: according to formula
Calculate after disinthibiting in red green passage subprovince and respond R
r3(x, y; σ
u), green red passage subprovince responds R after disinthibiting
g3(x, y; σ
u), blue yellow passage subprovince responds R after disinthibiting
b3(x, y; σ
u), wherein, (p, q), for dropping on the point outside Unit Nei Chu center, subprovince, MAX represents and gets higher value in both;
Step 6: according to formula
Calculate red green passage periphery and suppress R
r2(x, y; σ
s), green red passage periphery suppresses R
g2(x, y; σ
s), blue yellow passage periphery suppresses R
b2(x, y; σ
s), wherein, (p, q) is for dropping on the point in the Surround of periphery;
Step 7: according to formula
Calculate after district of Hong Lv channel center suppresses and respond R
r1(x, y; σ
c), district of Lv Hong channel center responds R after suppressing
g1(x, y; σ
c), district of Lan Huang channel center responds R after suppressing
b1(x, y; σ
c), wherein, (p, q) is for dropping on the point in the Center of center; Get R
r1(x, y; σ
c), R
g1(x, y; σ
c), R
b1(x, y; σ
c) as the new red, green, blue component of central pixel point (x, y);
Step 8: after Mobility Center pixel (x, y) travels through full figure, to the new red component I of all pixels of image
r, green component I
g, blue component I
b, yellow color component I
ysue for peace respectively, excited rejection ratio K adds 1, and difference iteration on red green, green red, blue yellow antagonism passage, repeats step 4 to 8;
The condition of above-mentioned iteration termination is: often take turns after iteration, subchannel check this passage corresponding color component and, if after when its first derivative and second dervative are all less than the value preset, all passages stop iteration, export as red component I with red green passage
r, green red passage exports as green component I
g, blue yellow passage exports as blue component I
bthe inclined coloured image of synthesizing colourless.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410190874.8A CN103957395B (en) | 2014-05-07 | 2014-05-07 | There is the color constancy method of adaptive ability |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410190874.8A CN103957395B (en) | 2014-05-07 | 2014-05-07 | There is the color constancy method of adaptive ability |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103957395A CN103957395A (en) | 2014-07-30 |
CN103957395B true CN103957395B (en) | 2015-12-09 |
Family
ID=51334608
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410190874.8A Active CN103957395B (en) | 2014-05-07 | 2014-05-07 | There is the color constancy method of adaptive ability |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103957395B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105931206B (en) * | 2016-05-10 | 2019-06-28 | 深圳市和天创科技有限公司 | A kind of color image definition enhancing method of color constancy |
CN106204500B (en) * | 2016-07-28 | 2018-10-16 | 电子科技大学 | A method of realizing that different cameral shooting Same Scene color of image remains unchanged |
CN108537852B (en) * | 2018-04-17 | 2020-07-07 | 四川大学 | Self-adaptive color constancy method based on image local contrast |
CN109274950B (en) * | 2018-11-09 | 2021-05-07 | 北京旷视科技有限公司 | Image processing method and device and electronic equipment |
CN109856133B (en) * | 2019-01-29 | 2021-06-22 | 深圳市象形字科技股份有限公司 | Test paper detection method utilizing multiple illumination intensities and multiple color illumination |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101674490A (en) * | 2009-09-23 | 2010-03-17 | 电子科技大学 | Color image color constant method based on retina vision mechanism |
CN102306384A (en) * | 2011-08-26 | 2012-01-04 | 华南理工大学 | Color constancy processing method based on single image |
CN102509272A (en) * | 2011-11-21 | 2012-06-20 | 武汉大学 | Color image enhancement method based on color constancy |
-
2014
- 2014-05-07 CN CN201410190874.8A patent/CN103957395B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101674490A (en) * | 2009-09-23 | 2010-03-17 | 电子科技大学 | Color image color constant method based on retina vision mechanism |
CN102306384A (en) * | 2011-08-26 | 2012-01-04 | 华南理工大学 | Color constancy processing method based on single image |
CN102509272A (en) * | 2011-11-21 | 2012-06-20 | 武汉大学 | Color image enhancement method based on color constancy |
Non-Patent Citations (1)
Title |
---|
基于图像倒数框架和非负数稀疏编码的颜色恒常计算方法;杜馨瑜等;《电子学报》;20120115;第40卷(第1期);179-183页 * |
Also Published As
Publication number | Publication date |
---|---|
CN103957395A (en) | 2014-07-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103957395B (en) | There is the color constancy method of adaptive ability | |
WO2020224518A1 (en) | Self-adaptive infrared image defringe algorithm based on local median histogram | |
CN106101697B (en) | Approach for detecting image sharpness, device and test equipment | |
CN110991266B (en) | Binocular face living body detection method and device | |
CN109068025B (en) | Lens shadow correction method and system and electronic equipment | |
CN106951869B (en) | A kind of living body verification method and equipment | |
US10145790B2 (en) | Image processing apparatus, image processing method, image capturing device and storage medium | |
CN106056559A (en) | Dark-channel-prior-method-based non-uniform-light-field underwater target detection image enhancement method | |
CN103955900B (en) | Image defogging method based on biological vision mechanism | |
CN102300100B (en) | White balance control device and method | |
CN109274948B (en) | Image color correction method, device, storage medium and computer equipment | |
CN103258332A (en) | Moving object detection method resisting illumination variation | |
CN105701773B (en) | A kind of method and device of quick processing image | |
CN102288884A (en) | External insulation discharging detecting method based on ultraviolet light spots | |
CN112399163B (en) | Image sensing device and automatic white balance method | |
CN103974053A (en) | Automatic white balance correction method based on grey dot extraction | |
CN101674490B (en) | Color image color constant method based on retina vision mechanism | |
CN104392425A (en) | Face based automatic contrast adjusting image enhancing method | |
CN109949250B (en) | Image processing method and device | |
CN104282013B (en) | A kind of image processing method and device for foreground target detection | |
CN111179293B (en) | Bionic contour detection method based on color and gray level feature fusion | |
US20200134807A1 (en) | Idiosyncrasy sensing system and idiosyncrasy sensing method | |
CN113091915A (en) | Automatic body temperature information acquisition system based on face recognition | |
CN108805144A (en) | Shell hole recognition methods based on morphology correction and system, indication of shots equipment | |
CN108600640A (en) | Gesture identification, color white balance adjustment and exposure regulating method and device in image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |