CN107180439A - A kind of colour cast feature extraction and colour cast detection method based on Lab chrominance spaces - Google Patents
A kind of colour cast feature extraction and colour cast detection method based on Lab chrominance spaces Download PDFInfo
- Publication number
- CN107180439A CN107180439A CN201610559667.4A CN201610559667A CN107180439A CN 107180439 A CN107180439 A CN 107180439A CN 201610559667 A CN201610559667 A CN 201610559667A CN 107180439 A CN107180439 A CN 107180439A
- Authority
- CN
- China
- Prior art keywords
- colour cast
- histogram
- feature
- image
- nno
- 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
Landscapes
- Image Analysis (AREA)
- Facsimile Image Signal Circuits (AREA)
Abstract
The invention discloses a kind of colour cast feature extraction based on Lab chrominance spaces and colour cast detection method, colour cast feature extracting method includes:One) colour cast feature h in the height direction, the colour cast feature for defining NNO regions colour cast changing features rate, defining luminance channel, two are defined) extract the colour cast feature of colour cast feature h and luminance channel;Colour cast detection method includes:1) equivalence circle feature u and σ of the chroma histogram in ab planes are calculated according to formula;2) " preliminary colour cast detection " flow is performed;3) " no colour cast image is detected again " flow etc. is performed.The present invention considers the peak Distribution characteristic of chromatic graph histogram in the height direction more fully hereinafter, and further contemplate the Assembled distribution characteristic of changing rule of the colour cast feature in original image and NNO regions and colour cast feature in luminance channel, deficiency of the existing method in terms of the definition and extraction of colour cast feature is compensate for, colour cast accuracy of detection is improved.
Description
Technical field
The present invention relates to the colour cast detection technique field of coloured image, more particularly to a kind of colour cast feature extraction and colour cast inspection
The method of survey.
Background technology
The mankind observe universal time, and reflection of the optic nerve to color, shape, surface tactile sensation and minutia gradually weakens.
Also, compared with word and sound, coloured image has more abundant intension and more powerful expressive ability.Therefore, color
It is to obtain the mostly important feature clue of objective world information, is also the key factor of direct weighing device image quality.However,
When the photosensitive coefficient of scene environment light source, object reflection characteristic in itself or collecting device changes, imaging device is remembered
The color value of objects in images under record deviates with respect to itself color, and entire image produces color offset phenomenon.This not only shadow
The visual effect of image is rung, while influence can be produced on subsequent treatments such as image segmentation, target identifications, therefore colour cast detection and color
Correction is often Color Image Processing and the indispensable link of field of machine vision partially.
Colour cast detection the most classical is the class method based on color constancy.The general process of such method is first
Estimate the illumination of scene, the input picture shot under unknown illumination is then transformed into by von Kries models by standard illumination
Under, so as to reach the purpose of colour cast detection and correction.Wherein, White Patch algorithms assume to be permanently present one in scene in vain
Color table face, estimation light source is used as using the maximum of RGB color passage in image.Grey World algorithms assume color in scene
Information enriches enough and the reflection of all physical surfaces is no color differnece (i.e. grey), by asking RGB color passage in image
Averagely it is worth to estimation light source.Shades of Grey algorithms are assumed whole scene is still after nonlinear inverible transform
No color differnece, replaces simply averaging, have ignored the local correlations between pixel using Minkowski-norm distances.Based on shellfish
The color constancy algorithm of this reasoning of leaf is by setting up the model between surface reflectivity and image irradiation, from color of image distribution
Posterior probability in estimate the illumination of image.Color constancy algorithm based on neutral net or SVM, then be utilized respectively many
The mapping model that layer neutral net or SVR are set up between scene illumination distribution and color of image distribution, and according to the model prediction
The illumination of new input picture.But, such colour cast detection method based on color constancy is generally only adapted to and meets specific
The application environment of suppositive scenario, if it is assumed that condition can not be met, then algorithm adaptability is difficult to get a desired effect.
In recent years, it is contemplated that have certain vector correlation between the RGB color passage of coloured image, a class is based on
The colour cast detection algorithm of Lab chroma histograms is widely applied.F.Gasparini etc. has carried out complete to such algorithm
Description.First, by the conversion of RGB to Lab color spaces, the luminance component of L * channel and the chromatic component of ab passages are obtained.So
Afterwards, using the method based on circle of equal value and NNO regions, the distribution character of chroma histogram is counted, colour cast feature is realized
It is quantitative to calculate and colour cast detection.Similarly, Li Feng and Chen etc. judges color with the two-dimensional chromaticity distribution characteristics of Lab color spaces
Partially with non-colour cast image, and propose that a kind of Gaussian Mixture Clustering Model makes a distinction to essential colour cast and true colour cast.Relative to
Colour cast detection method based on color constancy, such method is without suppositive scenario, with more preferable scene adaptability, and energy
Essential colour cast and true colour cast are made a distinction.
But, in the colour cast feature extraction based on equivalence circle, chroma histogram is substantially only considered in ab planes
Overall projected position information, this causes such method still to seem coarse and not in terms of the definition and extraction of colour cast feature
Foot, so that directly results in existing Lab colour casts detection algorithm still has relatively low colour cast accuracy of detection.
The content of the invention
In view of this, examined it is an object of the invention to provide a kind of colour cast feature extraction based on Lab chrominance spaces and colour cast
Survey method, it considers the peak Distribution characteristic of chromatic graph histogram in the height direction more fully hereinafter, and further contemplates
Changing rule of the colour cast feature in original image and NNO regions and colour cast feature luminance channel Assembled distribution characteristic, with
The existing colour cast detection algorithm based on Lab chroma histograms is solved to still have not in terms of the definition and extraction of colour cast feature
Foot, so as to directly result in the relatively low technical problem of colour cast accuracy of detection.
Colour cast feature extracting method of the invention based on Lab chrominance spaces, comprises the following steps:
One) colour cast feature is defined
1) the two-dimensional chromaticity histogram distribution function for defining a, b passage is:
H=F (a, b)
Wherein, a, b ∈ [- 128,127] are respectively the chromatic value of a, b Color Channel, and H is to correspond to chromatic value in image
The pixel number of (a, b), distribution function F () is a three-dimensional spatial distribution function;
2) defining the colour cast feature h of chroma histogram in the height direction is:
Wherein,And hσThe average and variance of chroma histogram in the height direction are represented respectively;
3) NNO regions colour cast changing features rate is defined:
Wherein, uNNOAnd σNNOThe equivalence circle feature for being NNO area colorimetrics histogram in ab planes, hNNOFor NNO regions color
The colour cast feature of histogram in the height direction is spent, then ucr、σcrAnd hcrThe equivalence circle feature and height of chroma histogram are represented respectively
Spend relative change rate of the feature between original image and NNO area images;
4) the brightness histogram data envelopment fitting algorithm based on gauss of distribution function, defines the plan of matched curve and envelope curve
The half width c of conjunction degree R and fitting function as luminance channel colour cast feature,
The gauss of distribution function is:
Wherein, a, b and c represent peak value size, peak center position and the half width information of Gaussian Profile respectively, and x and y divide
Biao Shi not independent variable and functional value;
Two) colour cast feature is extracted
1) colour cast feature h is extracted, is comprised the following steps:
A) coloured image of different sizes is normalized to size identical original image I to be measured1(i,j);
B) the chroma histogram height matrix H of normalization image is calculated0:
Wherein, n=256 represents the chrominance levels of a, b passage, hij=F (i-129, j-129) represents two-dimensional chromaticity histogram
Correspond to chromatic component a=i-129, b=j-129 pixel number in distribution function;
C) to H0Operation is filtered, to eliminate the influence of picture noise and the smaller element of peak value in chroma histogram:
H1=H0,if hij< T1M1then hij=0
Wherein, H1For filtered chroma histogram height matrix, T1For filtering threshold, M1For height matrix H0Average:
D) average of chroma histogram in the height direction is calculatedWith variance hσ:
Wherein, p is H1In 8 neighborhood connected regions number, Ωc, c=1 ..., p represents c-th of locally connected domain, ScFor
The connection area in the region,For the local height average in the region;
2) the colour cast feature of luminance channel is extracted, is comprised the following steps:
A) the L * channel histogram matrix N of normalization image is calculated0,
N0=[n0 n1 … n100]
Wherein, ni=f (i), i=0~100 represent the pixel number corresponding to brightness L=i, and f () is brightness Nogata
Figure function;
B) to brightness histogram N0Operation is filtered, to eliminate picture noise and the smaller member of peak value in brightness histogram
The influence of element, obtains filtered L * component histogram matrix N1:
N1=N0,if ni< T2S2then ni=0
Wherein, S2=max (ni) it is brightness histogram matrix N0In greatest member, T2For filtering threshold;
C) use step 4) in gauss of distribution function, based on least-squares algorithm, to brightness histogram N after filtering1Bag
Network data (x=i, y=ni), i=0~100 carry out curve fitting, the L * component histogram matrix N after being fitted2:
Wherein,The fitting result corresponding to brightness L=i is represented, g () is step 4) in
The gauss of distribution function of description, its parameter a, b and c are now tried to achieve by fitting algorithm;
D) according to the L * component histogram matrix N after filtering and after fitting1And N2, both fittings are calculated using the coefficient of determination
Degree R:
Wherein, i=0~100 represent L * channel brightness degree, niRepresent histogram matrix N after filtering1In correspond to brightness L
=i sample value,Represent histogram matrix N after fitting2In correspond to brightness L=i match value,For N1Middle all elements
Average, R ∈ [0,1] illustrate joint approximation ratio of the match value to sample value, and R is higher closer to 1 fitting precision.
The invention also discloses a kind of colour cast detection method based on Lab chrominance spaces, comprise the following steps:
Step 1):
According to formula (1):Formula (2):C=(ua,ub),With
Formula (3):Calculate equivalence circle feature u and σ of the chroma histogram in ab planes;F (a, b) in formula (1)
It is chroma histogram distribution function, k=a, b represents to be integrated on a axles or b axles, ukAnd σkRespectively chroma histogram is in k
Average and variance on axle;C is the center of circle of circle of equal value in formula (2), and σ is the radius of circle of equal value;
Step 2):
Perform " preliminary colour cast detection " flow:If meeting formula (4):(D > 10and Dσ> 0.6) or (Dσ> 1.5) bar
Part, image is classified as " colour cast ", goes to step 5);Otherwise, image is classified as " no colour cast ", goes to step 3);
Step 3):
Perform " no colour cast image is detected again " flow:According to formula (6):Criterion, image is divided into " colour cast ", " no colour cast " and " None- identified " three
Class;For " None- identified " image, go to step 4);For " colour cast " image, go to step 5).
Step 4):
Perform " None- identified image is detected again " flow:
1) feature extraction algorithm according to claim 1, calculates the colour cast of chroma histogram in the height direction
Feature h;
2) feature extraction algorithm according to claim 1, calculates changing features rate of the chrominance channel in NNO regions
ucr、σcrAnd hcr;
3) following criterion is used, None- identified image is detected again:
Step 5):
Perform " colour cast image is detected again " flow:
1) feature extraction algorithm according to claim 1, calculates colour cast the feature R and c of luminance channel;
2) following criterion is used, colour cast image is detected again:
Beneficial effects of the present invention:
Colour cast feature extraction and colour cast detection method of the present invention based on Lab chrominance spaces, are detected in classical Lab colour casts
On the basis of algorithm, the peak Distribution characteristic of chromatic graph histogram in the height direction is considered more fully hereinafter, and further consider
To changing rule of the colour cast feature in original image and NNO regions and colour cast feature luminance channel Assembled distribution characteristic,
Deficiency of the existing method in terms of the definition and extraction of colour cast feature is compensate for, colour cast accuracy of detection is improved.
Brief description of the drawings
Fig. 1 is the colour cast detection method flow chart based on Lab chrominance spaces in embodiment.
Embodiment
The invention will be further described below.
Colour cast detection algorithm based on Lab chroma histograms, its core concept is the distribution character from image color information
Set out, investigate the vector correlation between color.For the distribution character of quantitative analysis chroma histogram, classical Lab colour casts inspection
Method of determining and calculating realizes colour cast feature extraction and colour cast detection by introducing circle of equal value and NNO regions.Classical Lab colour casts detection algorithm
Flow is as follows:
Step 1:To coloured image to be detected, carry out RGB to Lab color space conversions, obtain luminance component image L with
Chromatic component a, b.
Step 2:The histogram of chromatic component is counted, following theoretical foundation is can be found that according to its distribution character:
Chroma histogram without colour cast image should show as multiple discrete peak values, and most of peak values should be distributed in neutral point (a=0,
B=0 around);And the chroma histogram for having colour cast image shows as the peak value concentrated comprising zero or one, and the peak value deviates
Neutral point.
Step 3:In order to which the intensity and its peak Distribution of quantitative description chroma histogram and the distance of neutral point are closed
System, introduces circle of equal value and chroma histogram feature is calculated.
Wherein, F (a, b) is chroma histogram distribution function, and k=a, b represents to be integrated on a axles or b axles, ukAnd σk
Respectively average and variance of the chroma histogram on k axles.Then according to formula (1), the center of circle C and radius sigma of circle of equal value are calculated:
C=(ua,ub),
On this basis, it is defined as follows circle colour cast feature of equal value:
Wherein, u is distance of the center of circle to neutral point (a=0, b=0), and D arrives the distance of neutral point, D for circle outside of equal valueσ
Represent that equivalence circle deviates the degree of neutral point.DσValue is bigger, shows that the chroma histogram deviation neutral point of the image is more serious
Or its peaks cluster is stronger, i.e. colour cast degree is more serious.
Step 4:Equivalence circle feature to chroma histogram is analyzed, when meeting following condition,
(D > 10and Dσ> 0.6) or (Dσ> 1.5) (4)
It is to deviate neutral point and peaks cluster to think chroma histogram, is tentatively classified as " colour cast image ";Otherwise, tentatively return
For " no colour cast image ".
Step 5:Further consider following theoretical foundation:No color differnece surface (the grey under standard white light in image scene
The neutral gray area on surface, i.e. coloured image) color that incident light shines in scene can be reflected completely, therefore by scene
The analysis in NNO regions, just can accurately estimate the illumination offset situation of image.The method for asking for image NNO regions is as follows:
And d non-orphaneds (5)
Wherein, L, a, b are respectively three components of the image in Lab color spaces, and d is colourity radius maximum, in order to anti-
Only noise jamming, limits d place pixels and each NNO area pixels point INNO(i, j) is non-orphaned point.
Step 6:Image for being tentatively classified as " no colour cast " in step 4, extracts its NNO area image and is examined again
Survey, the colour cast feature D of its NNO area image is solved using formula (1)-formula (3) identical methodσNNO, and perform following judgement:
Step 7:Image for being classified as " colour cast " in step 4 and step 6, classification knowledge is carried out using cluster learning algorithm
Not:If the region area comprising the dominant hue color such as ocean, blue sky, meadow in image scene reach the 40% of total image area with
On, it is classified as " essential colour cast ";Otherwise it is assumed that being " true colour cast ".
Obviously, the accuracy of Lab colour casts detection algorithm, dependent on the vector correlation to Lab Color-spatial distribution characteristics
Analysis and quantitative calculating.Therefore, present invention is generally directed to the colour cast feature and its feature extraction algorithm based on Lab chroma histograms
It is improved, on the basis of new colour cast feature and classics Lab colour cast detection algorithms, reintroduces complete colour cast detection improvement and calculate
Method.
Colour cast feature extracting method of the present embodiment based on Lab chrominance spaces comprises the following steps:
One) the two-dimensional chromaticity histogram distribution function for defining a, b passage is:
H=F (a, b) (7)
Wherein, a, b ∈ [- 128,127] are respectively the chromatic value of a, b Color Channel, and H is to correspond to chromatic value in image
The pixel number of (a, b), distribution function F () is a three-dimensional spatial distribution function.Detect and calculate in classical Lab colour casts
In method, circle of equal value is unique foundation that colour cast feature is extracted from chroma histogram distribution, although circle of equal value can be retouched preferably
The integral position (center of circle is to neutral point apart from u) and overall aggregation (radius sigma) equal distribution characteristic of chroma histogram are stated, still,
In the two-dimentional ab planes where by the histogrammic three-dimensional spatial distribution projection mapping of chromatic graph to circle of equal value during this, it is clear that lose
Histogram height information this important feature clue is lost.
Find under study for action, the information of chroma histogram in the height direction, colour cast image can be also reflected well and non-
The difference degree of colour cast image.Its theoretical foundation is:For without colour cast image, because various chromatic components are more rich in image
Richness, is presented more obvious dispersiveness, therefore the differing distribution of its dominant hue and other secondary tones in the height direction is smaller;
And for colour cast image, because obvious aggregation is presented in chromatic component, therefore its dominant hue and other secondary tones are in height
Distributional difference on direction is larger.
Two) for quantitative description and chroma histogram size distribution in the height direction and difference degree are calculated, this
Embodiment defines the colour cast feature h of chroma histogram in the height direction:
Wherein,And hσThe average and variance of chroma histogram in the height direction are represented respectively.
The specific algorithm for extracting colour cast feature h is described as follows:
Step 1:Coloured image of different sizes is normalized to size identical original image I to be measured1(i, j), so
Although the image after processing is there may be distortion from visual effect, the statistical property of its distribution of color does not become
Change.
Step 2:Calculate the chroma histogram height matrix H of normalization image0:
Wherein, n=256 represents the chrominance levels of a, b passage, hij=F (i-129, j-129) represents two-dimensional chromaticity histogram
Correspond to chromatic component a=i-129, b=j-129 pixel number in distribution function.
Step 3:To H0Operation is filtered, to eliminate the shadow of picture noise and the smaller element of peak value in chroma histogram
Ring:
H1=H0,if hij< T1M1then hij=0 (10)
Wherein, H1For filtered chroma histogram height matrix, T1For filtering threshold, M1For height matrix H0It is equal
Value:
Step 4:Calculate the average of chroma histogram in the height directionWith variance hσ:
Wherein, p is H1In 8 neighborhood connected regions number, Ωc, c=1 ..., p represents c-th of locally connected domain, ScFor
The connection area in the region,For the local height average in the region.
Three) in classical Lab colour casts detection algorithm, the two-dimensional chromaticity histogram equivalence circle of NNO area images is only accounted for
Parameter characteristic, not combine original image relevant parameter carry out changing features before and after quantitative analysis.Therefore, to preliminary
During being determined as that the NNO regions of " no colour cast image " are detected again, it is easy to occur misjudgment phenomenon.
Find in our study, colour cast feature of the chroma histogram on disk of equal value and short transverse, if
Quantitative analysis is carried out to changing features rule with reference to original image and NNO area images, can preferably reflect colour cast image and Fei Se
The difference degree of inclined image.Its theoretical foundation is:Reduce amplitude for the radius of circle of equal value without colour cast image, its NNO region
Larger and closer to neutral point, the changing features amplitude of the chroma histogram in its NNO region in the height direction is smaller;And it is right
In colour cast image, the radius of circle of equal value reduction amplitude in its NNO region is smaller and farther away from neutral point, the colourity in its NNO region
The changing features amplitude of histogram in the height direction is larger.In order to which quantitative description and calculating colour cast feature are in original image and NNO
Changing rule between area image, the NNO regions colour cast changing features rate that the present embodiment is further defined as follows:
Wherein, uNNOAnd σNNOThe equivalence circle feature for being NNO area colorimetrics histogram in ab planes, hNNOFor NNO regions
The colour cast feature of chroma histogram in the height direction, then ucr、σcrAnd hcrRespectively represent chroma histogram equivalence circle feature and
Relative change rate of the altitude feature between original image and NNO area images.
Four) in classical Lab colour casts detection algorithm, two-dimensional chromaticity component a, b distribution character is only considered, is not considered
Luminance component L channel characteristic.Find in our study, the L * component in Lab chrominance spaces can preferably reflect idiochromatic
Difference degree between inclined and true colour cast:For essential colour cast image, its L * channel histogram shows zonal unimodal poly-
Collection distribution;For true colour cast image, its L * channel histogram shows more uniform discrete distribution.
Consider following gauss of distribution function:
Wherein, a, b and c represent peak value size, peak center position and the half width information of Gaussian Profile respectively, and x and y divide
Biao Shi not independent variable and functional value.This gauss of distribution function also shows zonal unimodal Assembled distribution characteristic, and half-breadth
Spend c smaller, represent that the Gaussian Profile is more assembled.
For quantitative description and the calculating histogrammic distribution character of L * channel, the present embodiment proposes a kind of based on Gaussian Profile
Brightness histogram data envelopment fitting algorithm, and define matched curve and envelope curve fitting degree R and fitting function half
Width c as luminance channel colour cast feature.Obviously, fitting degree higher (R is bigger) and region clustering is better (c is smaller),
Represent that the histogrammic unimodal aggregation of L * channel is more obvious, corresponding coloured image may more belong to essential colour cast.
The specific algorithm for extracting the colour cast feature of luminance channel is described as follows:
Step 1:Calculate the L * channel histogram matrix N of normalization image0,
N0=[n0 n1 … n100] (15)
Wherein, ni=f (i), i=0~100 represent the pixel number corresponding to brightness L=i, and f () is brightness Nogata
Figure function.
Step 2:To brightness histogram N0Operation is filtered, it is smaller to eliminate picture noise and peak value in brightness histogram
The influence of element, obtains filtered L * component histogram matrix N1:
N1=N0,if ni< T2S2then ni=0 (16)
Wherein, S2=max (ni) it is brightness histogram matrix N0In greatest member, T2For filtering threshold, the present embodiment takes
T2=0.2.
Step 3:The gauss of distribution function described using formula (14), based on least-squares algorithm, to brightness Nogata after filtering
Scheme N1Envelope data (x=i, y=ni), i=0~100 carry out curve fitting, the L * component histogram matrix after being fitted
N2:
Wherein,The fitting result corresponding to brightness L=i is represented, g () is formula (14) description
Gauss of distribution function, its parameter a, b and c are now tried to achieve by fitting algorithm.
Step 4:According to the L * component histogram matrix N after filtering and after fitting1And N2, calculate both using the coefficient of determination
Fitting degree R:
Wherein, i=0~100 represent L * channel brightness degree, niRepresent histogram matrix N after filtering1In correspond to brightness L
=i sample value,Represent histogram matrix N after fitting2In correspond to brightness L=i match value,For N1Middle all elements
Average, R ∈ [0,1] illustrate joint approximation ratio of the match value to sample value, and R is higher closer to 1 fitting precision.
Colour cast detection method of the present embodiment based on Lab chrominance spaces comprises the following steps:
Step 1:According to formula (1)-(3), equivalence circle feature u and σ of the chroma histogram in ab planes are calculated.
Step 2:Perform " preliminary colour cast detection " flow:If meeting formula (4) condition, image is classified as " colour cast ", goes to step 5;
Otherwise, image is classified as " no colour cast ", goes to step 3.
Step 3:Perform " no colour cast image is detected again " flow:According to formula (6) criterion, image is divided into " colour cast ", " colourless
Partially " and " None- identified " three class;For " None- identified " image, 4 are gone to step;For " colour cast " image, 5 are gone to step.
Step 4:Perform " None- identified image is detected again " flow:
1) according to foregoing colour cast feature h extraction algorithm, the colour cast feature h of chroma histogram in the height direction is calculated;
2) according to the feature extraction algorithm of the colour cast feature of foregoing luminance channel, spy of the chrominance channel in NNO regions is calculated
Levy rate of change ucr、σcrAnd hcr;
3) following criterion is used, None- identified image is detected again:
Step 5:Perform " colour cast image is detected again " flow:
1) according to the feature extraction algorithm of the colour cast feature of foregoing luminance channel, calculate luminance channel colour cast feature R and
c;
2) following criterion is used, colour cast image is detected again:
In order to verify the validity of colour cast feature proposed by the invention in the present embodiment, construct by 480 cromograms
As the test data set of composition.Wherein, colour cast and the numeral of non-colour cast image are respectively 191 and 289.It is first in this experiment
First using obtaining improved colour cast feature in the present embodiment based on the colour cast feature extracting method of Lab chrominance spaces.Then, make
Colour cast detection is carried out with the colour cast detection method based on Lab chrominance spaces in the present embodiment.For classical Lab colour casts detection algorithm
Step 6) in unrecognized image, following processing is done to obtain higher accuracy:
It was found from the testing result listed by table one, colour cast feature proposed by the invention and feature extracting method are all feasible
's;In addition, compared with the classical colour cast detection algorithm based on Lab, feature and detection method proposed by the present invention have higher
Accuracy.
The testing result of all test images of table one
Finally illustrate, the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted, although with reference to compared with
The present invention is described in detail good embodiment, it will be understood by those within the art that, can be to skill of the invention
Art scheme is modified or equivalent substitution, and without departing from the objective and scope of technical solution of the present invention, it all should cover at this
Among the right of invention.
Claims (2)
1. a kind of colour cast feature extracting method based on Lab chrominance spaces, it is characterised in that:Comprise the following steps:
One) colour cast feature is defined
1) the two-dimensional chromaticity histogram distribution function for defining a, b passage is:
H=F (a, b)
Wherein, a, b ∈ [- 128,127] are respectively the chromatic value of a, b Color Channel, and H is to correspond to chromatic value (a, b) in image
Pixel number, distribution function F () is a three-dimensional spatial distribution function;
2) defining the colour cast feature h of chroma histogram in the height direction is:
Wherein,And hσThe average and variance of chroma histogram in the height direction are represented respectively;
3) NNO regions colour cast changing features rate is defined:
Wherein, uNNOAnd σNNOThe equivalence circle feature for being NNO area colorimetrics histogram in ab planes, hNNOIt is straight for NNO area colorimetrics
Side schemes colour cast feature in the height direction, then ucr、σcrAnd hcrRepresent that the equivalence circle feature and height of chroma histogram are special respectively
Levy the relative change rate between original image and NNO area images;
4) the brightness histogram data envelopment fitting algorithm based on gauss of distribution function, defines the fitting journey of matched curve and envelope curve
The half width c of R and fitting function is spent as the colour cast feature of luminance channel,
The gauss of distribution function is:
Wherein, a, b and c represent peak value size, peak center position and the half width information of Gaussian Profile, x and y difference tables respectively
Show independent variable and functional value;
Two) colour cast feature is extracted
1) colour cast feature h is extracted, is comprised the following steps:
A) coloured image of different sizes is normalized to size identical original image I to be measured1(i,j);
B) the chroma histogram height matrix H of normalization image is calculated0:
Wherein, n=256 represents the chrominance levels of a, b passage, hij=F (i-129, j-129) represents two-dimensional chromaticity histogram distribution
Correspond to chromatic component a=i-129, b=j-129 pixel number in function;
C) to H0Operation is filtered, to eliminate the influence of picture noise and the smaller element of peak value in chroma histogram:
H1=H0,if hij< T1M1 then hij=0
Wherein, H1For filtered chroma histogram height matrix, T1For filtering threshold, M1For height matrix H0Average:
D) average of chroma histogram in the height direction is calculatedWith variance hσ:
Wherein, p is H1In 8 neighborhood connected regions number, Ωc, c=1 ..., p represents c-th of locally connected domain, ScFor the area
The connection area in domain,For the local height average in the region;
2) the colour cast feature of luminance channel is extracted, is comprised the following steps:
A) the L * channel histogram matrix N of normalization image is calculated0,
N0=[n0 n1 … n100]
Wherein, ni=f (i), i=0~100 represent the pixel number corresponding to brightness L=i, and f () is brightness histogram letter
Number;
B) to brightness histogram N0Operation is filtered, to eliminate the shadow of picture noise and the smaller element of peak value in brightness histogram
Ring, obtain filtered L * component histogram matrix N1:
N1=N0,if ni< T2S2 then ni=0
Wherein, S2=max (ni) it is brightness histogram matrix N0In greatest member, T2For filtering threshold;
C) use step 4) in gauss of distribution function, based on least-squares algorithm, to brightness histogram N after filtering1Envelope number
According to (x=i, y=ni), i=0~100 carry out curve fitting, the L * component histogram matrix N after being fitted2:
Wherein,The fitting result corresponding to brightness L=i is represented, g () is step 4) described in
Gauss of distribution function, its parameter a, b and c are now tried to achieve by fitting algorithm;
D) according to the L * component histogram matrix N after filtering and after fitting1And N2, both fitting degrees are calculated using the coefficient of determination
R:
Wherein, i=0~100 represent L * channel brightness degree, niRepresent histogram matrix N after filtering1In correspond to brightness L=i
Sample value,Represent histogram matrix N after fitting2In correspond to brightness L=i match value,For N1The average of middle all elements,
R ∈ [0,1] illustrate joint approximation ratio of the match value to sample value, and R is higher closer to 1 fitting precision.
2. a kind of colour cast detection method based on Lab chrominance spaces, it is characterised in that:Comprise the following steps:
Step 1):
According to formula (1):Formula (2):And formula
(3):Calculate equivalence circle feature u and σ of the chroma histogram in ab planes;F (a, b) is color in formula (1)
Histogram distribution function is spent, k=a, b represents to be integrated on a axles or b axles, ukAnd σkRespectively chroma histogram is on k axles
Average and variance;C is the center of circle of circle of equal value in formula (2), and σ is the radius of circle of equal value;
Step 2):
Perform " preliminary colour cast detection " flow:If meeting formula (4):(D > 10and Dσ> 0.6) or (Dσ> 1.5) condition,
Image is classified as " colour cast ", goes to step 5);Otherwise, image is classified as " no colour cast ", goes to step 3);
Step 3):
Perform " no colour cast image is detected again " flow:According to formula (6):
Criterion, image is divided into " colour cast ", " no colour cast " and " None- identified " three class;For " None- identified " image, go to step 4);For
5) " colour cast " image, go to step.
Step 4):
Perform " None- identified image is detected again " flow:
1) feature extraction algorithm according to claim 1, calculates the colour cast feature of chroma histogram in the height direction
h;
2) feature extraction algorithm according to claim 1, calculates changing features rate u of the chrominance channel in NNO regionscr、
σcrAnd hcr;
3) following criterion is used, None- identified image is detected again:
Step 5):
Perform " colour cast image is detected again " flow:
1) feature extraction algorithm according to claim 1, calculates colour cast the feature R and c of luminance channel;
2) following criterion is used, colour cast image is detected again:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610559667.4A CN107180439B (en) | 2016-07-15 | 2016-07-15 | Color cast characteristic extraction and color cast detection method based on Lab chromaticity space |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610559667.4A CN107180439B (en) | 2016-07-15 | 2016-07-15 | Color cast characteristic extraction and color cast detection method based on Lab chromaticity space |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107180439A true CN107180439A (en) | 2017-09-19 |
CN107180439B CN107180439B (en) | 2020-02-04 |
Family
ID=59830522
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610559667.4A Active CN107180439B (en) | 2016-07-15 | 2016-07-15 | Color cast characteristic extraction and color cast detection method based on Lab chromaticity space |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107180439B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110807739A (en) * | 2019-09-17 | 2020-02-18 | 中国科学院自动化研究所 | Image color feature processing method, system and device for target detection and storage medium |
CN111091532A (en) * | 2019-10-30 | 2020-05-01 | 中国资源卫星应用中心 | Remote sensing image color evaluation method and system based on multilayer perceptron |
CN111724396A (en) * | 2020-06-17 | 2020-09-29 | 泰康保险集团股份有限公司 | Image segmentation method and device, computer-readable storage medium and electronic device |
CN112184838A (en) * | 2020-10-09 | 2021-01-05 | 哈尔滨工程大学 | Multi-background camouflage pattern dominant color extraction method based on color correlation |
CN114136959A (en) * | 2021-11-27 | 2022-03-04 | 北京中医药大学 | Urine color quantification method and classification standard |
CN114322833A (en) * | 2021-12-31 | 2022-04-12 | 中国科学院长春光学精密机械与物理研究所 | White light scanning interference three-dimensional reconstruction method based on pseudo Wigner-Ville distribution |
CN117524090A (en) * | 2023-12-08 | 2024-02-06 | 广州卓奥科技有限公司 | LED display color self-calibration method and system based on artificial intelligence |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110135200A1 (en) * | 2009-12-04 | 2011-06-09 | Chao-Ho Chen | Method for determining if an input image is a foggy image, method for determining a foggy level of an input image and cleaning method for foggy images |
CN103402117A (en) * | 2013-08-06 | 2013-11-20 | 夏东 | Method for detecting color cast of video image based on Lab chrominance space |
CN103500457A (en) * | 2013-10-23 | 2014-01-08 | 武汉东智科技有限公司 | Method of color cast detection of video image |
CN104168478A (en) * | 2014-07-29 | 2014-11-26 | 银江股份有限公司 | Video image off-color detection method based on Lab space and correlation function |
CN104202596A (en) * | 2014-09-17 | 2014-12-10 | 西安电子科技大学 | Image color-cast detection method and system applied to intelligent terminal |
-
2016
- 2016-07-15 CN CN201610559667.4A patent/CN107180439B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110135200A1 (en) * | 2009-12-04 | 2011-06-09 | Chao-Ho Chen | Method for determining if an input image is a foggy image, method for determining a foggy level of an input image and cleaning method for foggy images |
CN103402117A (en) * | 2013-08-06 | 2013-11-20 | 夏东 | Method for detecting color cast of video image based on Lab chrominance space |
CN103500457A (en) * | 2013-10-23 | 2014-01-08 | 武汉东智科技有限公司 | Method of color cast detection of video image |
CN104168478A (en) * | 2014-07-29 | 2014-11-26 | 银江股份有限公司 | Video image off-color detection method based on Lab space and correlation function |
CN104202596A (en) * | 2014-09-17 | 2014-12-10 | 西安电子科技大学 | Image color-cast detection method and system applied to intelligent terminal |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110807739A (en) * | 2019-09-17 | 2020-02-18 | 中国科学院自动化研究所 | Image color feature processing method, system and device for target detection and storage medium |
CN111091532A (en) * | 2019-10-30 | 2020-05-01 | 中国资源卫星应用中心 | Remote sensing image color evaluation method and system based on multilayer perceptron |
CN111724396A (en) * | 2020-06-17 | 2020-09-29 | 泰康保险集团股份有限公司 | Image segmentation method and device, computer-readable storage medium and electronic device |
CN112184838A (en) * | 2020-10-09 | 2021-01-05 | 哈尔滨工程大学 | Multi-background camouflage pattern dominant color extraction method based on color correlation |
CN114136959A (en) * | 2021-11-27 | 2022-03-04 | 北京中医药大学 | Urine color quantification method and classification standard |
CN114322833A (en) * | 2021-12-31 | 2022-04-12 | 中国科学院长春光学精密机械与物理研究所 | White light scanning interference three-dimensional reconstruction method based on pseudo Wigner-Ville distribution |
CN114322833B (en) * | 2021-12-31 | 2022-09-06 | 中国科学院长春光学精密机械与物理研究所 | White light scanning interference three-dimensional reconstruction method based on pseudo Wigner-Ville distribution |
CN117524090A (en) * | 2023-12-08 | 2024-02-06 | 广州卓奥科技有限公司 | LED display color self-calibration method and system based on artificial intelligence |
CN117524090B (en) * | 2023-12-08 | 2024-06-21 | 广州卓奥科技有限公司 | LED display color self-calibration method and system based on artificial intelligence |
Also Published As
Publication number | Publication date |
---|---|
CN107180439B (en) | 2020-02-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107180439A (en) | A kind of colour cast feature extraction and colour cast detection method based on Lab chrominance spaces | |
Zheng et al. | Image segmentation based on adaptive K-means algorithm | |
CN103035013B (en) | A kind of precise motion shadow detection method based on multi-feature fusion | |
CN110046673A (en) | No reference tone mapping graph image quality evaluation method based on multi-feature fusion | |
Huang et al. | Detection of and compensation for shadows in colored urban aerial images | |
CN102385753B (en) | Illumination-classification-based adaptive image segmentation method | |
CN109871845B (en) | Certificate image extraction method and terminal equipment | |
CN103914699A (en) | Automatic lip gloss image enhancement method based on color space | |
CN112598761B (en) | Collaborative design and intelligent recommendation method for color matching of textile and clothing | |
Zakir et al. | Road sign segmentation based on colour spaces: A Comparative Study | |
Hussain et al. | Color constancy algorithm for mixed-illuminant scene images | |
CN109741285B (en) | Method and system for constructing underwater image data set | |
CN106503638A (en) | For the image procossing of colour recognition, vehicle color identification method and system | |
CN108711160B (en) | Target segmentation method based on HSI (high speed input/output) enhanced model | |
Tian et al. | Outdoor shadow detection by combining tricolor attenuation and intensity | |
Hussain et al. | Color constancy for uniform and non-uniform illuminant using image texture | |
CN109685772A (en) | It is a kind of based on registration distortion indicate without referring to stereo image quality appraisal procedure | |
CN114331946A (en) | Image data processing method, device and medium | |
Zhang et al. | Atmospheric perspective effect enhancement of landscape photographs through depth-aware contrast manipulation | |
CN115457551A (en) | Leaf damage identification method suitable for small sample condition | |
CN110070480A (en) | A kind of analogy method of underwater optics image | |
Panetta et al. | Novel multi-color transfer algorithms and quality measure | |
Vršnak et al. | Framework for illumination estimation and segmentation in multi-illuminant scenes | |
Naccari et al. | Natural scenes classification for color enhancement | |
Huang et al. | Image dehazing based on robust sparse representation |
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 |