CN104202596B - A kind of image colour cast detection method and system being applied to intelligent terminal - Google Patents

Abstract

The invention discloses a kind of the image colour cast detection method and the system that are applied to intelligent terminal, image acquisition; Space transforming; The histogram of computed image under ab two-dimensional color coordinate and the parameter of equivalence circle; Solve the NNO region of source images and converted images information and corresponding Circle Parameters of equal value thereof respectively; Quantization threshold, reduced parameter, judges image colour cast situation, carries out colour cast Images Classification, or non-colour cast image detects again; Through repeated detection, image is divided into colour cast image, essential colour cast image, true colour cast image, without colour cast image, cannot detected image.The present invention has good applicability, high colour cast Detection accuracy and reliability; Reduce algorithm time and complexity, there is good real-time performance, be suitable for intelligent terminal and take pictures the real-time detection of image.

Description

A kind of image colour cast detection method and system being applied to intelligent terminal

Technical field

The invention belongs to digital image processing techniques field, particularly relate to a kind of the image colour cast detection method and the system that are applied to intelligent terminal.

Background technology

One of most important information that color comprises as image is the key character of direct weighing device image quality.Effect diagram is the semaphore stored in digital imaging apparatus internal image transducer as the key factor of quality of colour, and its content not only depends on the surface color of collected object, also will be subject to extraneous light conditions at that time.Therefore when imaging device as digital camera, scanner operationally, the impact being subject to factors due to its inner sensitive component causes finally obtaining the error existed between the color of color digital image and the realistic colour of real scene body surface to a certain degree, i.e. so-called colour cast.

In non-reference picture quality appraisement field, colour cast is an important measurement index.Current traditional algorithm realizes according to color constancy the detection of colour cast.Classical algorithm has max_RGB algorithm, GreyWorld algorithm, Retinex algorithm etc.There is the shortcoming that accuracy is lower, algorithm is complicated, time complexity is high in these algorithms! Be not suitable for and intelligent terminal occasion that computational resource limited higher to requirement of real-time.

Summary of the invention

The object of the embodiment of the present invention is to provide a kind of the image colour cast detection method and the system that are applied to intelligent terminal, and being intended to solve does not have a kind of problem of more exactly image being carried out to colour cast detection method at present.

The embodiment of the present invention is achieved in that a kind of image colour cast detection method being applied to intelligent terminal, and this image colour cast detection method being applied to intelligent terminal comprises:

The first step, in Lab space, the histogram of computed image under ab two-dimensional color coordinate and the parameter of equivalence circle;

Second step, solves the NNO region of source images and converted images information and corresponding Circle Parameters of equal value respectively; Quantization threshold, reduced parameter, judges image colour cast situation, carries out colour cast Images Classification, or non-colour cast image detects again;

3rd step, through repeated detection, is divided into colour cast image by image, essential colour cast image, true colour cast image, without colour cast image, and cannot detected image.

Further, needed to obtain a secondary RGB image Img_sur as image to be detected by intelligent terminal of taking pictures before the parameter of the histogram of first step computed image under ab two-dimensional color coordinate and equivalence circle; Source images Img_sur is transformed into Lab chrominance space by rgb space, obtains converted images Img_Lab and the view data on L, a, b component thereof.

Further, colour cast Images Classification comprises the following steps: calculate luminance component histogram; The number of times occurred according to pixel and interval accounting judge colour cast type for the first time; Image colour cast type is finally determined: essential colour cast, true colour cast by NNO region and source images two-dimensional histogram Circle Parameters of equal value.

Further, the concrete grammar of colour cast Images Classification comprises:

Step one, through type (1), the histogram distribution feature of quantitative analysis Img_Lab image under ab two-dimensional color coordinate, and calculate circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σparameter;

$\left\{\begin{array}{c}{u}_k=\underset{k}{\∫}\mathrm{kF}(a,b)\mathrm{dk},\\ {\mathrm{\σ}}_k^2=\underset{k}{\∫}{(u_k-k)}^{2}F(a,b)\mathrm{dk}\end{array}\right.,k=a,b---\left(1\right)$

$\left\{\begin{array}{c}C=(u_a,u_b)\\ \mathrm{\σ}=\sqrt{{{\mathrm{\σ}}_a}^2+{{\mathrm{\σ}}_b}^2}\end{array}\right.---\left(2\right)$

$\left\{\begin{array}{c}u=\sqrt{{u_a}^2+{u_b}^2}\\ D=u-\mathrm{\σ}\\ D_{\mathrm{\σ}}=D/\mathrm{\σ}\end{array}\right.---\left(3\right)$

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis: D _σbe worth larger, show that this image histogram departs from neutral axis more serious;

Step 2, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and be temporarily colour cast image by graphic collection, otherwise preliminary identification image is non-colour cast image, performs step step 3;

(D>10andD _σ>0.6)or(D _σ>1.5)(4)

Step 3, solves the NNO region of image; Theoretical foundation: the no color differnece surface in image scene can reflect the color of incident illumination in scene completely.As long as therefore extract the gray face in scene, then by the colour cast situation of these gray faces, the drift condition of the illumination of image just can be estimated accurately; Method is as follows: if Img_NNO (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, and d is the colourity radius maximum of testing image Img_sur in Lab chrominance space; Meanwhile, for preventing the noise spot in image for the interference of NNO extracted region, restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point.

Further, it is supplementing and correcting last result that non-colour cast image detects again, and step is: the two-dimensional histogram equivalence circle generating source images and NNO area image respectively, quantization threshold, reduced parameter, output detections result: essential colour cast, true colour cast, without colour cast.

Further, the concrete grammar that non-colour cast image detects again comprises:

Step one, solves the corresponding circle D of equal value of its two-dimensional histogram to result images _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image or non-colour cast image;

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, the isoparametric amplitude of variation of two-dimensional histogram equivalence circle σ, u of normal non-colour cast image source image and NNO area image is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of source images, and this characteristic conforms is normally without the color characteristic of colour cast image; Meanwhile, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, and namely the NNO area image of colour cast image has the color characteristic of colour cast equally; So it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, and setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO<-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO>0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image;

Step 2, to being judged to be that the image of colour cast image is classified for the first time:

By the histogram distribution information utilizing image at the L component of Lab chrominance space, image is classified first; When picture material is obvious essential colour cast, the histogram distribution of the L component of Lab chrominance space presents zonal Assembled distribution; And picture material is when being true colour cast, the histogram distribution of the L component of Lab chrominance space is rendered as more uniform discrete distribution; In the following way the L component of image is processed:

When certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel; Judge L histogram of component after treatment, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component;

When non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast; When being greater than 80, then the judgement of products for further;

To having essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, and the change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in NNO region comes further to judge; Color constancy; If the distortion of image generation colour cast, then NNO region also can embody colour cast feature; If colour cast distortion does not occur image, NNO region should show the feature of normal picture, and use formula (6) to classify to colour cast image equally, arranging of threshold value adjusts according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively.

Further, this image colour cast detection method being applied to intelligent terminal specifically comprises the following steps:

Step one, obtains a secondary RGB image Img_sur as image to be detected by intelligent terminal capable of taking pictures;

Step 2, is transformed into Lab chrominance space by source images Img_sur by rgb space, obtains converted images Img_Lab and the view data on L, a, b component;

Step 3, through type (1), the histogram distribution feature of quantitative analysis Img_Lab image under ab two-dimensional color coordinate, and calculate circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σparameter;

$\left\{\begin{array}{c}{u}_k=\underset{k}{\&Integral;}\mathrm{kF}(a,b)\mathrm{dk},\\ {\mathrm{\&sigma;}}_k^2=\underset{k}{\&Integral;}{(u_k-k)}^{2}F(a,b)\mathrm{dk}\end{array}\right.,k=a,b---\left(1\right)$

$\left\{\begin{array}{c}C=(u_a,u_b)\\ \mathrm{\&sigma;}=\sqrt{{{\mathrm{\&sigma;}}_a}^2+{{\mathrm{\&sigma;}}_b}^2}\end{array}\right.---\left(2\right)$

$\left\{\begin{array}{c}u=\sqrt{{u_a}^2+{u_b}^2}\\ D=u-\mathrm{\&sigma;}\\ D_{\mathrm{\&sigma;}}=D/\mathrm{\&sigma;}\end{array}\right.---\left(3\right)$

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis: D _σbe worth larger, show that this image histogram departs from neutral axis more serious;

Step 4, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and is temporarily colour cast image by graphic collection, performs step 7; Otherwise preliminary identification image is non-colour cast image, performs step 5;

(D>10andD _σ>0.6)or(D _σ>1.5)(4)

Step 5, solve NNO (nearneutralobjects) region of image, theoretical foundation: the no color differnece surface in image scene can reflect the color of incident illumination in scene completely, by the colour cast situation of these gray faces, estimate the drift condition of the illumination of image accurately, method is as follows: if Img_NNO (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, d is the colourity radius maximum of testing image Img_sur in Lab chrominance space, simultaneously, for preventing the noise spot in image for the interference of NNO extracted region, restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point;

Step 6, solves the corresponding circle D of equal value of its two-dimensional histogram to the result images of step 5 _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image or non-colour cast image;

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, there is good parameter role of delegate, the two-dimensional histogram equivalence circle σ of normal non-colour cast image source image and NNO area image, the isoparametric amplitude of variation of u is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of source images, this characteristic conforms is normally without the color characteristic of colour cast image, simultaneously, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, namely the NNO area image of colour cast image has the color characteristic of colour cast equally, so, it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, and setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO<-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO>0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image;

Step 7, to being judged to be that the image of colour cast image is classified for the first time:

By the histogram distribution information utilizing image at the L component of Lab chrominance space, classify first to image, when picture material is obvious essential colour cast, the histogram distribution of the L component of Lab chrominance space presents zonal Assembled distribution; And picture material is when being true colour cast, the histogram distribution of the L component of Lab chrominance space is rendered as more uniform discrete distribution, processes in the following way to the L component of image:

When certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel;

Judge L histogram of component after treatment, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component:

When non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast;

When being greater than 80, then the judgement of products for further;

To having essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in their NNO region comes further to judge, color constancy, if the distortion of image generation colour cast, then NNO region also can embody colour cast feature; If colour cast distortion does not occur image, NNO region should show the feature of normal picture, uses formula (6) to classify to colour cast image equally, and wherein, arranging of threshold value adjusts according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively;

Step 8, is embedded into the program of the method in mobile terminal capable of taking pictures, detects in real time the picture that equipment obtains, to obtain the evaluation to this hardware performance in color.

Another object of the embodiment of the present invention is to provide a kind of image colour cast detection system being applied to intelligent terminal, it is characterized in that, this image colour cast detection system being applied to intelligent terminal comprises:

By RGB image acquisition; Space transforming; The histogram of computed image under ab two-dimensional color coordinate and the parameter of equivalence circle; Solve NNO (nearneutralobjects) region of source images and converted images information and corresponding Circle Parameters of equal value thereof respectively; Quantization threshold, reduced parameter, judges image colour cast situation, carries out colour cast Images Classification, or " non-colour cast image " detects again; Through repeated detection, image is divided into colour cast image, essential colour cast image, true colour cast image, without colour cast image, cannot detected image.

The present invention can not be subject to the limitation of scene or priori, and compared to existing technologies, improves the precision of detection, has general applicability simultaneously, improves accuracy rate and the reliability of colour cast detection; And the present invention has used conversion and the histogram of color space neatly, NNO region, some knowledge of circle of equal value, simplify operand, there is good real-time performance, be suitable for intelligent terminal and take pictures the real-time detection of image, also can be used for the performance evaluating intelligent terminal image-taking system.

Because current most of graph image storehouse is not all set up specially for the image library set of colour cast research, therefore, the performance comparison situation of algorithm and other algorithms herein comprehensively cannot be analyzed.In order to verify the accuracy of innovatory algorithm effect herein, I has collected a large amount of colour casts, normal picture (192 width experimental image altogether from network, wherein 20 width essence colour cast images, the true colour cast image of 30 width, 109 width normal pictures and the 33 width image that cannot judge), and former Lab chrominance space colour cast detection algorithm is performed respectively and algorithm after improving to this image collection, and the experiment detection statistics result of two algorithms is contrasted, the Detection accuracy of main contrast two algorithm and False Rate, testing result contrast situation is as shown in the table:

As seen from the above table, innovatory algorithm is relative to former algorithm all increasing in classification and the detection accuracy without the context of detection again of colour cast image at colour cast image.Meanwhile, because innovatory algorithm does not need the evaluation and classification algorithm by other, make algorithm in whole flow process be closed, establish one's own system, can be used in fast detecting and evaluate image color information application scenario in.

Accompanying drawing explanation

Fig. 1 is the image colour cast detection method flow chart being applied to intelligent terminal that the embodiment of the present invention provides;

Fig. 2 is the flow chart being applied to the image colour cast detection method embodiment of intelligent terminal that the embodiment of the present invention provides;

Fig. 3 is the colour cast image classification module schematic flow sheet that the embodiment of the present invention provides;

Fig. 4 is the non-colour cast image detection module schematic flow sheet again that the embodiment of the present invention provides;

Fig. 5 is the embedding Android device interface schematic diagram that the embodiment of the present invention provides;

Fig. 6 is the embedding Android device data manipulation flow chart that the embodiment of the present invention provides.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with embodiment, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

Below in conjunction with drawings and the specific embodiments, application principle of the present invention is further described.

As shown in Figure 1, the image colour cast detection method being applied to intelligent terminal of the embodiment of the present invention comprises the following steps:

S101:RGB image acquisition: obtained by terminal imaging device;

S102: space transforming: RGB image is transformed into Lab space;

S103: the histogram of computed image under ab two-dimensional color coordinate and the parameter of equivalence circle;

S104: solve NNO (nearneutralobjects) region of source images and converted images information and corresponding Circle Parameters of equal value thereof respectively;

S105: quantization threshold, reduced parameter, judges image colour cast situation, carries out colour cast Images Classification, or " non-colour cast image " detects again;

S106: through repeated detection, is divided into colour cast image by image, essential colour cast image, true colour cast image, without colour cast image, and cannot detected image.

Concrete steps of the present invention are:

Step one, obtains a secondary RGB image Img_sur as image to be detected by intelligent terminal capable of taking pictures;

Step 2, is transformed into Lab chrominance space by source images Img_sur by rgb space, obtains converted images Img_Lab and the view data on L, a, b component thereof;

Step 3, through type (1), the histogram distribution feature of quantitative analysis Img_Lab image under ab two-dimensional color coordinate, and calculate circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σetc. parameter;

$\left\{\begin{array}{c}{u}_k=\underset{k}{\&Integral;}\mathrm{kF}(a,b)\mathrm{dk},\\ {\mathrm{\&sigma;}}_k^2=\underset{k}{\&Integral;}{(u_k-k)}^{2}F(a,b)\mathrm{dk}\end{array}\right.,k=a,b---\left(1\right)$

$\left\{\begin{array}{c}C=(u_a,u_b)\\ \mathrm{\&sigma;}=\sqrt{{{\mathrm{\&sigma;}}_a}^2+{{\mathrm{\&sigma;}}_b}^2}\end{array}\right.---\left(2\right)$

$\left\{\begin{array}{c}u=\sqrt{{u_a}^2+{u_b}^2}\\ D=u-\mathrm{\&sigma;}\\ D_{\mathrm{\&sigma;}}=D/\mathrm{\&sigma;}\end{array}\right.---\left(3\right)$

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis: D _σbe worth larger, show that this image histogram departs from neutral axis more serious;

Step 4, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and be temporarily " colour cast image " by graphic collection, execution step 7; Otherwise preliminary identification image is " non-colour cast image ", performs step 5;

(D>10andD _σ>0.6)or(D _σ>1.5)(4)

Step 5, solve NNO (nearneutralobjects) region of image, theoretical foundation: the no color differnece surface (gray face under standard white light in image scene, the i.e. neutral gray area of coloured image) color of incident illumination in scene can be reflected completely, as long as therefore extract the gray face in scene, then by the colour cast situation of these gray faces, just can estimate the drift condition of the illumination of image accurately, method is as follows: if Img_NNO is (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, d is the colourity radius maximum of testing image Img_sur in Lab chrominance space, simultaneously, for preventing the noise spot in image for the interference of NNO extracted region, restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point;

Step 6, solves the corresponding circle D of equal value of its two-dimensional histogram to the result images of step 5 _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image (non-colour cast image);

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, there is good parameter role of delegate, the two-dimensional histogram equivalence circle σ of normal non-colour cast image source image and NNO area image, the isoparametric amplitude of variation of u is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of source images, this characteristic conforms is normally without the color characteristic of colour cast image, simultaneously, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, namely the NNO area image of colour cast image has the color characteristic of colour cast equally, so, it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, based on the analysis and synthesis of abundant experimental results, setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO<-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO>0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image; Other situations then regard as the image type that algorithm cannot process;

Step 7, to being judged to be that the image of " colour cast image " is classified for the first time:

1, the histogram distribution information will image being utilized at the L component of Lab chrominance space, classifies first to image, and when picture material is obvious essential colour cast, the histogram distribution of the L component of its Lab chrominance space presents zonal Assembled distribution; And picture material is when being true colour cast, the histogram distribution of the L component of its Lab chrominance space is rendered as more uniform discrete distribution, through a large amount of experiments, processes in the following way to the L component of image:

1.1, when certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel;

1.2, judge the L histogram of component after 1.1 step process, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component:

1.2.1, when non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast;

1.2.2, when being greater than 80, then the judgement of products for further;

2, to those, there is essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in their NNO region comes further to judge, the principle of this step is: color constancy, if the distortion of image generation colour cast, then its NNO region also can embody colour cast feature; If colour cast distortion does not occur image, its NNO region should show the feature of normal picture, through localized experimental verification, uses formula (6) to classify to colour cast image equally, and wherein, arranging of threshold value can adjust according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively; ;

Step 8, is embedded into the program of the method in mobile terminal capable of taking pictures, detects in real time the picture that equipment obtains, to obtain the evaluation to this hardware performance in color;

Specific embodiments of the invention:

As shown in Figure 2, the embodiment of the present invention by colour cast image classification module (Fig. 3), non-colour cast image module detects (Fig. 4) two parts composition again, the algorithm idea of these two modules is embedded into Android device (intelligent Android system mobile phone), design user interface as shown in Figure 5, on the main interface of mobile phone A pp, places three control: ImageView, for showing the image that camera obtains, it is namely exactly source images to be evaluated; TextView, provides the evaluation result of image, and whether the image namely obtained by this mobile phone exists colour cast, and is the colour cast of which kind of type; Button, arranges monitoring event onClick, when after trigger event, the camera-enabled that calling mobile phone is built-in, be defaulted as and call post-positioned pick-up head (if having front-facing camera), detects after getting image, and result feedback is shown to TextView, the operating process of whole data flow as shown in Figure 6, is divided into five module: Button, equipment camera, obtain image, colour cast algorithm, evaluation result, is described in detail the operating process according to data flow below:

Button: in Android exploitation, Button is one of control the most frequently used in various UI, user can trigger sequence of events by touching it, one does not have the Button of click event without any meaning, on stream, the ID that Button mono-usually will be given unique or key, called after button herein, can for it arranges association attributes in XML file, such as wide, high, mainly in the onCreat of Activity, obtain button example by findViewById, then button is bound to the monitor onClickListener of a click event, by button click invocation facility camera, ,

Equipment camera: Google provides the API calling built-in camera, only needs to obtain corresponding authority, just can, by the camera on built-in camera API Calls mobile phone, realize taking pictures and obtaining image;

Obtain image: obtaining image by calling camera, needing man-machine interactively here to realize image taking;

Colour cast algorithm: the implementation language of algorithm based on the system embedded by smart machine, and in Android platform, realizes this algorithm by Java language, and process is as follows:

Step one, is transformed into Lab chrominance space by source images Img_sur by rgb space, obtains the image Img_Lab after changing and the view data on L, a, b component thereof;

Step 2, through type (1), obtains the histogram distribution feature of Img_Lab image under ab two-dimensional color coordinate, and calculates circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σetc. parameter;

$\left\{\begin{array}{c}{u}_k=\underset{k}{\&Integral;}\mathrm{kF}(a,b)\mathrm{dk},\\ {\mathrm{\&sigma;}}_k^2=\underset{k}{\&Integral;}{(u_k-k)}^{2}F(a,b)\mathrm{dk}\end{array}\right.,k=a,b---\left(1\right)$

$\left\{\begin{array}{c}C=(u_a,u_b)\\ \mathrm{\&sigma;}=\sqrt{{{\mathrm{\&sigma;}}_a}^2+{{\mathrm{\&sigma;}}_b}^2}\end{array}\right.---\left(2\right)$

$\left\{\begin{array}{c}u=\sqrt{{u_a}^2+{u_b}^2}\\ D=u-\mathrm{\&sigma;}\\ D_{\mathrm{\&sigma;}}=D/\mathrm{\&sigma;}\end{array}\right.---\left(3\right)$

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis;

Step 3, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and be temporarily " colour cast image " by graphic collection, execution step 7; Otherwise preliminary identification image is " non-colour cast image ", performs step 5;

(D>10andD _σ>0.6)or(D _σ>1.5)(4)

Step 4, solves the NNO region of image, if Img_NNO (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, d is the colourity radius maximum of testing image Img_sur in Lab chrominance space, simultaneously, for preventing the noise spot in image for the interference of NNO extracted region, restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point;

Step 5, solves the corresponding circle D of equal value of its two-dimensional histogram to result images _σthe parameter of NNO, and the parameters contrasting source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carry out secondary judgement from the variation tendency of both parameters to non-agglomerated image (non-colour cast image);

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO<-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO>0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image; Other situations then regard as the image type that algorithm cannot process;

Step 6, to being judged to be that the image of " colour cast image " is classified for the first time:

1, utilize image at the histogram distribution information of the L component of Lab chrominance space, image classified first, in the following way the L component of image is processed:

1.1, when certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel;

1.2, judge the L histogram of component after i step process, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component:

1.2.1, when non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast; ;

1.2.1, when being greater than 80, then the judgement of products for further;

2, to those, there is essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, the change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in NNO region comes further to judge.Experiment can be supplied because colour cast detects image library ununified at present, all sample image are originated equal network retrieval, through localized experimental verification, use formula (6) to classify to colour cast image equally, wherein, arranging of threshold value can adjust according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively;

Evaluation result: the result feedback of colour cast algorithm is shown to TextView.

The real-time that the present invention is good, repeatedly judges when colour cast image detects, and makes testing result more reliable, accurate; More accurate in colour cast Images Classification.By a large amount of experiments, as far as possible objectively in quantization algorithm about the select permeability of threshold value--image belongs to the threshold value of colour cast or non-colour cast, and colour cast image belongs to the threshold value of true colour cast or essential colour cast; More accurate in the context of detection again of non-colour cast image.The method analyzes situation of change before and after the corresponding round parameter of equal value of two-dimensional histogram further, and propose situation about distributing according to the L component information of image first colour cast image is divided into true colour cast and essential colour cast, can more effectively supplement primary testing result, the truth of effective evaluation image colour cast; While the accuracy of algorithm is improved, remain the features such as the real-time of former algorithm dexterously, and accurately to colour cast Images Classification, the performance of imaging product can be evaluated better, choose reasonable consumptive material.

Due to its lightweight, easy realization, real-time is good, accuracy high can be embedded into multiple mobile terminal capable of taking pictures, not only comprise the terminal of Android system, also comprise iOS, the operating systems such as windows, directly can be reflected the performance of the acquisition image of intelligent terminal capable of taking pictures by the detection and classification of colour cast; Not only may be used for the end product that directly can obtain digital picture, but also may be used for evaluating color static imaging (laser printing, xerography) product, adopt certain measure, by paper image digitlization, thus for good and bad degree, the particularly quality of compatible consumptive material that light source information obtains in perception color static imaging objectively (laser printing, xerography) product imaging scene.For color printer, the formula of carbon dust or the electrophotographic image forming performance of printer, all may produce comparatively serious impact to the color of image.Range of application widely.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (8)

1. be applied to an image colour cast detection method for intelligent terminal, it is characterized in that, this image colour cast detection method being applied to intelligent terminal comprises:

The first step, in Lab space, the histogram of computed image under ab two-dimensional color coordinate and the parameter of equivalence circle;

Second step, solve the NNO region of source images and converted images information and corresponding Circle Parameters of equal value, quantization threshold and reduced parameter respectively, judge image colour cast situation, carry out colour cast Images Classification, or non-colour cast image detects and solves two-dimensional histogram corresponding of equal value circle D to result images _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using following formula as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image or non-colour cast image;

σ _cr＝(σ-σ _NNO)/σ，u _cr＝(u-u _cr)/u

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, the isoparametric amplitude of variation of two-dimensional histogram equivalence circle σ, u of normal non-colour cast image source image and NNO area image is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of original image, and this characteristic conforms is normally without the color characteristic of colour cast image; Meanwhile, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, and namely the NNO area image of colour cast image has the color characteristic of colour cast equally; So it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, and parameter threshold is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO <-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO > 0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image;

3rd step, through being not more than 3 detections, is divided into colour cast image by image, without colour cast image, and cannot detected image.

2. be applied to the image colour cast detection method of intelligent terminal as described in claim 1, it is characterized in that, needed to obtain a secondary RGB image Img_sur as image to be detected by intelligent terminal of taking pictures before the parameter of the histogram of first step computed image under ab two-dimensional color coordinate and equivalence circle; Source images Img_sur is transformed into Lab chrominance space by rgb space, obtains converted images Img_Lab and the image brightness on L, a, b component thereof and chroma data; Wherein to be transformed into Lab chrominance space formula as follows for rgb space:

3. be applied to the image colour cast detection method of intelligent terminal as described in claim 1, it is characterized in that, colour cast Images Classification comprises the following steps: calculate luminance component histogram; The number of times occurred according to pixel and interval institute accounting judge colour cast type for the first time; Image colour cast type is finally determined: essential colour cast, true colour cast by NNO region and source images two-dimensional histogram Circle Parameters of equal value.

4. be applied to the image colour cast detection method of intelligent terminal as described in claim 3, it is characterized in that, the concrete grammar of colour cast Images Classification comprises:

Step one, through type (1), analyzes the histogram distribution feature of Img_Lab image under ab two-dimensional color coordinate, and calculates circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σparameter;

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis: D _σbe worth larger, show that this image histogram departs from neutral axis more serious;

Step 2, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and is temporarily colour cast image by graphic collection, otherwise, tentatively assert that image is non-colour cast image, perform step 3;

(D＞10andD _σ＞0.6)or(D _σ＞1.5)(4)

Step 3, solves the NNO region of image; Theoretical foundation: the no color differnece surface in image scene can reflect the color of incident illumination in scene completely; Extract the gray face in scene, by the colour cast situation of gray face, draw the drift condition of the illumination of image; Method is as follows: if Img_NNO (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, and d is the colourity radius maximum of testing image Img_sur in Lab chrominance space; Restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point.

5. be applied to the image colour cast detection method of intelligent terminal as described in claim 1, it is characterized in that, it is supplementing and correcting last result that non-colour cast image detects again, step is: the two-dimensional histogram equivalence circle generating source images and NNO area image respectively, quantization threshold and reduced parameter, output detections result: essential colour cast, true colour cast, without colour cast.

6. be applied to the image colour cast detection method of intelligent terminal as described in claim 5, it is characterized in that, the concrete grammar that non-colour cast image detects again comprises:

Step one, solves the corresponding circle D of equal value of two-dimensional histogram to result images _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image or non-colour cast image;

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, the isoparametric amplitude of variation of two-dimensional histogram equivalence circle σ, u of normal non-colour cast image source image and NNO area image is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of original image, and this characteristic conforms is normally without the color characteristic of colour cast image; Meanwhile, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, and namely the NNO area image of colour cast image has the color characteristic of colour cast equally; So it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, and setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO <-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO > 0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image;

Step 2, to being judged to be that the image of colour cast image is classified for the first time:

By the histogram distribution information utilizing image at the L component of Lab chrominance space, image is classified first; When picture material is obvious essential colour cast, the histogram distribution of the L component of Lab chrominance space presents zonal Assembled distribution; And picture material is when being true colour cast, the histogram distribution of the L component of Lab chrominance space is rendered as more uniform discrete distribution; In the following way the L component of image is processed:

When certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel; Judge L histogram of component after treatment, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component;

When non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast; When being greater than 80%, then the judgement of products for further;

To having essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, and the change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in NNO region comes further to judge; If the distortion of image generation colour cast, then NNO region also can embody colour cast feature; If colour cast distortion does not occur image, NNO region should show the feature of normal picture, and use formula (6) to classify to colour cast image equally, arranging of threshold value adjusts according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively.

7. be applied to the image colour cast detection method of intelligent terminal as described in claim 1, it is characterized in that, this image colour cast detection method being applied to intelligent terminal specifically comprises the following steps:

Step one, take a secondary coloured image Img_sur by the intelligent terminal possessing camera function, namely RGB image is image to be detected;

Step 2, is transformed into Lab chrominance space by source images Img_sur by rgb space, obtains converted images Img_Lab and the view data on L, a, b component;

Step 3, through type (1), the histogram distribution feature of quantitative analysis Img_Lab image under ab two-dimensional color coordinate, and calculate circle E of equal value according to formula (2), (3) _qcenter of circle C, radius sigma, u, D, D _σparameter;

In formula (3), u is the distance of center of circle C distance neutral axis (a=0, b=0) of this equivalence circle, and D is the distance of the outer lateral extent neutral axis of circle of equal value, D _σrepresent that this two-dimensional histogram equivalence circle departs from the degree of neutral axis: D _σbe worth larger, show that this image histogram departs from neutral axis more serious;

Step 4, judges image colour cast situation: for the first time when meeting formula (4), then think that the ab two dimensional surface histogram of image assembles, and is temporarily colour cast image by graphic collection, performs step 7; Otherwise preliminary identification image is non-colour cast image, performs step 5;

(D＞10andD _σ＞0.6)or(D _σ＞1.5)(4)

Step 5, solve the NNO region of image, theoretical foundation: the no color differnece surface in image scene can reflect the color of incident illumination in scene completely, by the colour cast situation of gray face, draw the drift condition of the illumination of image, method is as follows: if Img_NNO (i, j) pixel is NNO area pixel, then:

Wherein L, a, b are three component information of image at Lab chrominance space, d is the colourity radius maximum of testing image Img_sur in Lab chrominance space, restriction d place pixel and each NNO area pixel point Img_NNO (i, j) are non-isolated pixel point;

Step 6, solves the corresponding circle D of equal value of its two-dimensional histogram to the result images of step 5 _σthe parameter of NNO, the parameters of contrast source images equivalence circle, using formula (6) as the quantitative basis weighing image parameter change, carries out secondary judgement from the variation tendency of both parameters to non-agglomerated image or non-colour cast image;

σ _cr＝(σ-σ _NNO)/σ；u _cr＝(u-u _NNO)/u(6)

Wherein, σ _crand u _crrepresent two-dimensional image histogram radius of a circle σ of equal value and the amplitude of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively, effectively can extract the round situation of change of the equivalence of the two-dimensional histogram before and after NNO region by quantized image, there is good parameter role of delegate, the two-dimensional histogram equivalence circle σ of normal non-colour cast image source image and NNO area image, the isoparametric amplitude of variation of u is larger, namely the Circle Parameters of equal value of the NNO area image of normal non-colour cast image changes greatly than the Circle Parameters of equal value of original image, this characteristic conforms is normally without the color characteristic of colour cast image, simultaneously, σ, u parameter value variation amplitude of colour cast image before and after extraction NNO region is less, namely the NNO area image of colour cast image has the color characteristic of colour cast equally, so, it is effective for realizing the method for non-colour cast Images Classification by the change analyzing σ, u parameter value, and setup parameter threshold value is as follows:

As image NNO region Circle Parameters D of equal value _σduring NNO <-0.5, or, σ, u parameter extract the amplitude that changes before and after NNO region respectively more than 70% and 60% time, judge that image is normal non-colour cast image; As image NNO region Circle Parameters D of equal value _σduring NNO > 0.5, or, when the amplitude that σ, u parameter changes before and after extraction NNO region is respectively lower than 40% and 30%, judge that image is colour cast image;

Step 7, to being judged to be that the image of colour cast image is classified for the first time:

By the histogram distribution information utilizing image at the L component of Lab chrominance space, classify first to image, when picture material is obvious essential colour cast, the histogram distribution of the L component of Lab chrominance space presents zonal Assembled distribution; And picture material is when being true colour cast, the histogram distribution of the L component of Lab chrominance space is rendered as more uniform discrete distribution, processes in the following way to the L component of image:

When certain pixel occurrence number in L histogram of component lower than the highest pixel occurrence number of occurrence number 1% time, by the pixel count zero setting of this pixel;

Judge L histogram of component after treatment, the minimum pixel that pixel occurrence number is non-vanishing and maximum pixel across interval whether be less than 80% at the proportion of the whole pixel range of L component:

When non-zero number of times region that pixel is crossed over is less than or equal to whole pixel range 80%, judge that this image colour cast type is as essential colour cast;

When being greater than 80%, then the judgement of products for further;

To having essential colour cast feature, and picture material is not very single essential colour cast image and true colour cast image, change according to the relative source images of the parameters Circle Parameters of equal value of the two-dimensional histogram equivalence circle in their NNO region comes further to judge, if the distortion of image generation colour cast, then NNO region also can embody colour cast feature; If colour cast distortion does not occur image, NNO region should show the feature of normal picture, uses formula (6) to classify to colour cast image equally, and wherein, arranging of threshold value adjusts according to different scenes:

Wherein σ _crand u _crrepresent colour cast two-dimensional image histogram radius of a circle σ of equal value and the amplitude index of distance of center circle from distance u relative Self-variation before and after extraction NNO region of initial point respectively;

Step 8, the program of the method be embedded in mobile terminal capable of taking pictures, detect in real time the picture that equipment obtains, this program can feed back testing result, namely to the evaluation of this hardware performance in color.

8. be applied to an image colour cast detection system for intelligent terminal, it is characterized in that, this image colour cast detection system being applied to intelligent terminal comprises: colour cast image classification module, non-colour cast image module, intelligent Android system mobile phone;

On the main interface of intelligent Android system mobile phone App, placing three control: ImageView, for showing the image that camera obtains, is namely exactly source images to be evaluated; TextView, provides the evaluation result of image, and whether the image namely obtained by this mobile phone exists colour cast, and is the colour cast of which kind of type; Button, arrange monitoring event onClick, and when after trigger event, the camera-enabled that calling mobile phone is built-in, detects after getting image, and shown to TextView by result feedback.