CN101441763B - Multiple-colour tone image unity regulating method based on color transfer - Google Patents

Abstract

The invention discloses a method for uniformly adjusting a multicolor picture based on color transmission. The method comprises the following steps: 1, one picture with two colors is divided into two pictures, namely a reference picture and a picture to be adjusted; and the two pictures are converted to a 1 alpha beta color space; 2, the two pictures are divided through a K mean algorithm; and each partition region of the reference picture is sampled; 3, an expectation maximization algorithm is used to carry out probability partition on the two pictures respectively; and a mapping relation is established between each partition region of the reference picture and each partition region of the picture to be adjusted; 4, the picture to be adjusted is subjected to brightness zooming in order to keep consistency in brightness of the reference picture; and 5, a point of sampling points of the reference picture which is most matched with each point of the picture to be adjusted is searched out; and the color value of the most matched point of the reference picture is transmitted to the corresponding point of the picture to be adjusted. The method has small calculation amount, is convenient and rapid and solves the problem of color discontinuity in the prior multicolor picture adjustment method.

Description

Multi-level image based on the color transmission is unified method of adjustment

Technical field

The present invention relates to a kind of to unifying method of adjustment based on the multi-level image based on the color transmission of color transmission.

Background technology

There is an obvious defects in present remote sensing image: because the subimage of the zones of different in the original image possibly be to be taken under the difference moment and different weather conditions by the satellite of different model to obtain; Therefore when the later stage is spliced demonstration to these images; The problem that the adjacent image tone has big difference will appear; It is lofty visually to seem; The observer is done not feel like oneself and influence interpretation, can bring difficulty for analyzing and processing such as land resources statistics, physical environment monitoring, disaster assessment and the city planning of follow-up geodata, also can the impact analysis statistical accuracy.This problem also often occurs in most commercial GIS software, can see this problem often among the for example now very popular online generalized information system Google Earth by the release of Google company.This shows that the unified research of the tone of remote sensing images seems very necessary.

On present remote sensing image processing correlative study work mainly concentrates on feature extraction and classifies; The research of relevant even look seldom; The still traditional digital image processing method that is adopted in the practice at present, or utilize business software artificially such as Photoshop that image is carried out the tone adjustment.Briefly introduce the colored Enhancement Method of traditional digital picture below:

1) histogram is handled

Histogram is the statistics that the gray scale (colour) of piece image distributes, and has comprised abundant information.The histogram enhancement techniques is just with the foundation of histogram as conversion, makes histogram after the conversion become the shape of expectation.Histogram transformation commonly used has histogram equalization, histogram normalization and histogram coupling in the remote sensing image processing.

Histogram equalization is actually image is carried out non-linear stretching, redistributes gray scale (colour) value of image pixel, makes the pixel quantity in certain gray scale (colour) scope roughly the same.Histogram normalization is that the histogram transformation with image is the shape of normal distribution, and gray scale (colour) frequency distribution of image has the shape near the distribution of normal state if this is, just can think that this width of cloth image is fit to eye-observation.The histogram coupling is to be the histogram of certain designated shape or the histogram of a certain reference picture to the histogram transformation of original image; Adjust gray scale (colour) value of each pixel of source images then according to the histogram of known designated modality, obtain the image of a width of cloth histogram and reference histograms coupling at last.

2) wavelet transform process

Wavelet transformation be a kind of signal the time ask---yardstick (time---frequency) analytical approach; It has the characteristics of multiresolution analysis; And the time, frequently two territories all have the ability of characterization signal local feature; Be that a kind of window size immobilizes but its shape can change, the time-frequency localization analytical approach that time window and frequency window can change.Promptly have higher frequency resolution and lower temporal resolution, have at HFS and ask resolution and lower frequency resolution when higher, be described as the microscope of analytic signal in low frequency part.Wavelet transformation is applied to the figure image intensifying; Be that picture breakdown is size, the position component all different with direction; Before doing inverse transformation; Change the size of some coefficient in the wavelet transformed domain, so just can amplify interested component selectively and reduce unwanted component, thereby reach the purpose of realization figure image intensifying.

3) HIS adjustment

HSI is tone, saturation degree and intensity color model, and tone is an attribute of describing pure color, and saturation degree provides a kind of pure color by the tolerance of the degree of white light dilution, and brightness is subjective description, has embodied colourless intensity notion.This model can be in coloured image from the influence of chromatic information (color harmony saturation degree) the lining cancellation strength component that carries, the image processing method of describing based on colour for exploitation is a desirable instrument.The hue information of image can adopt several different methods that it is adjusted directly corresponding to the H component, to obtain the tone of expectation.

Summary of the invention

The objective of the invention is to overcome the deficiency of existing overall adjustment algorithm, the deficiency that the processing time is long provides a kind of multi-level image based on the color transmission to unify method of adjustment.

Unifying method of adjustment based on the multi-level image of color transmission may further comprise the steps:

The piece image that 1) will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space;

2) with the K mean algorithm two width of cloth images are cut apart, and each cut zone of reference picture is sampled;

3) with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust between each cut section of image and set up mapping relations;

4) treat the convergent-divergent that the adjustment image carries out brightness, it and reference brightness are kept consistency;

5) treat each point of adjustment image, in the sampled point of reference picture, seek and its point that matees most, the color value of match point in the reference picture is passed to the corresponding point of waiting to adjust in the image;

The described piece image that will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space, the steps include:

(1) piece image that has two kinds of tones is divided into two width of cloth images, is to exist the picture of two kinds of tones to carry out preliminary area dividing by the user to one, it is divided into two parts: wait to adjust picture and reference picture;

(2) if former picture uses is the RGB color space, then pass through following step, be converted into l α β color space, the steps include:

A) first step is to change into the XYZ tristimulus values to image from RGB, and wherein XYZ is the device independent space, and is as follows:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{ccc}0.5141& 0.3239& 0.1604\\ 0.2651& 0.6702& 0.0641\\ 0.0241& 0.1228& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

B) second step was to be converted into the LMS space from XYZ space, and is as follows:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3897& 0.6980& -0.0787\\ -0.2298& 1.1834& 0.0464\\ 0.0000& 0.0000& 1.0000\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]$

C) merge these two transformed matrixs, obtain transition matrix, obtain following matrix from the RGB color space to the LMS color space:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3811& 0.5783& 0.0402\\ 0.1967& 0.7244& 0.0782\\ 0.0241& 0.1288& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

D) calculate the logarithm value of LMS value:

L＝logL，M＝logM，S＝logS

Adopt following formula to carry out final conversion:

$\left[\begin{array}{c}l\\ \mathrm{\α}\\ \mathrm{\β}\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{3}}& 0& 0\\ 0& \frac{1}{\sqrt{6}}& 0\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& -2\\ 1& -1& 0\end{array}\right]\left[\begin{array}{c}L\\ M\\ S\end{array}\right]$

Describedly two width of cloth images are cut apart, and each cut zone of reference picture are carried out sampling step with the K mean algorithm:

(3) with the K mean algorithm two width of cloth images are cut apart, its step does;

E) point that will import is divided into K initial set, adopts disposable or all can heuristic cutting apart;

F) calculate the central point or the barycenter of each set;

G) utilize formula:

D _m=min|x _m-u _i| ², x _mBe the point of input, u _iBarycenter for each set

Each point is distributed to apart from its nearest central point;

H) recomputate the barycenter of each set, and utilize formula:

$V=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\underset{x_j\∈S_i}{\mathrm{\Σ}}{|x_j-u_i|}^2$

Calculate the difference of two squares in the cluster,, or after the step of iteration some, cut apart end when the value of its value less than certain user's appointment; Otherwise carrying out g and h step of iteration.

(4) each cut zone of reference picture is sampled, can sample by the user is manual, or the computer random sampling;

Describedly with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust and set up the mapping relations step between each cut section of image:

(5) use expectation-maximization algorithm that two width of cloth images are carried out the probability segmentation procedure respectively to be:

I) initialization: utilize the K average segmentation result of a last joint to calculate each mean value of areas u _iAnd standard deviation sigma _i

J) expectation: (x y) belongs to i Gauss model G to pixel I _i(i; u _i, σ _i) probability can pass through computes:

$i^P\mathrm{xy}=\frac{\exp (-\frac{{\left(I(x,y)u_i\right)}^2}{2{\mathrm{\σ}}_i^2})}{{\mathrm{\Σ}}_{j=1}^N\exp (-\frac{{\left(I(x,y)u_j\right)}^2}{2{\mathrm{\σ}}_j^2})}$

G _iRepresent regional i Gaussian distribution G _i(i; u _i, σ _i), _iP _Xy(x y) belongs to G to the pixel I of representative image I _iProbability, p _XyThe pixel I of representative image I (x, probability distribution estimation y), $p_{\mathrm{Xy}}^t=\left\{p_{\mathrm{Xy}}^t{}_i\right|i=1,2...\}.$

K) maximization: the Gaussian distribution G that recomputates regional i according to following formula _i(i; u _i, σ _i) average u _iAnd standard deviation sigma _i:

$u_i=\frac{1}{\underset{x,y}{{\mathrm{\Σ}}_i}{p}_{\mathrm{xy}}}\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}I(x,y)$

${\mathrm{\σ}}_i=\sqrt{\frac{\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}{(I(x,y)-u_i)}^2}{\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}}}$

L) repeat j, two steps of k up to convergence.

(6) to reference picture with wait to adjust between each cut section of image and set up mapping relations, step is:

M) we will set up the mapping function f () of each Gauss model

of waiting to adjust in the image some Gauss model in the reference picture; This mapping can be artificial the appointment; Also can be to accomplish automatically, but will satisfy the condition in the step 2 by algorithm;

N) suppose that image to be adjusted has two Gauss models

With

In luminance channel $u_{t_i}\≥u_{t_j},$ So, the average u ' after the mapping also should satisfy ${u_{t_i}}^{\′}\≥{u_{t_j}}^{\′}.$ In our algorithm, when

With

The brightness average the most approaching, and satisfy under the situation of monotonicity,

Be mapped to

The described adjustment image convergent-divergent that carries out brightness of treating makes it and the reference brightness step that keeps consistency:

(7) because the difference of two width of cloth images in overall brightness if directly adopt their brightness to compare, will cause very poor matching result, therefore, we need in advance the brightness of image to be changed, and they are consistent.We have adopted a kind of linear mapping to come the average and the standard deviation of the Luminance Distribution of two width of cloth images are shone upon:

$Y\′\left(p\right)=\frac{{\mathrm{\σ}}_s}{{\mathrm{\σ}}_t}(Y\left(p\right)-u_t)+u_s$

u _s, σ _s: reference brightness average and standard deviation

u _t, σ _t:: brightness average and the standard deviation of waiting to adjust image

Described each point of treating the adjustment image is sought in the sampled point of reference picture and its point that matees most, and the color value of match point in the reference picture is passed to the corresponding point step of waiting to adjust in the image:

(8) treat each point of adjusting image, in the sampled point of reference picture, seek and its point that matees most, the steps include:

O) for corresponding sampled point in each point of waiting to adjust image and the reference picture, at first whether judgement point and the sampled point in the reference picture waiting to adjust in the image belongs to the same area after the K average is cut apart, and promptly judges f (KMeansPartitionResult (P _t)) whether equal f (KMeansPartitionResult (P _s)).If two points are not at the same area, promptly above two functional values are unequal, then continue a pair of point relatively down; If at the same area, promptly above two functional values equate, then continue step 2 and mate;

P) with the brightness of point-to-point transmission and probability distribution as evaluation criterion, calculate P _tAnd P _sDistance, P _tAnd P _sDistance function definition as follows:

$\mathrm{dis}\tan \mathrm{ce}(p_s,p_t)=w_1|l_{p_s}-l_{p_t}|+w_2\left|\right|{\stackrel{\→}{P}}_{p_s}-{\stackrel{\→}{P}}_{p_t}\left|\right|$

Expression p _s, p _tLuminance difference

Expression point p _sProbability segmentation result vector

Some p after expression is shone upon through the zone _tProbability segmentation result vector

represents

and Euclidean distance between

Usually get w ₂∝ 10w ₁

Calculating waits to adjust distance between point and all sampled points in the reference picture in the image, and wherein that minimum sampled point of distance is optimal match point.

(9) color value with match point in the reference picture passes to the corresponding point of waiting to adjust in the image, the steps include:

Q) finding P _tMatch point P _sAfter, with P _sColouring information (being the value of α β passage) pass to P _t, keep P _tBrightness constant (being the value of l passage).

The beneficial effect that the present invention compared with prior art has:

Traditional length consuming time of the image adjusting method based on manual work, workload is big, and efficient is low; The research work of relevant tone adjustment mainly is to daily treatment of picture research, and be not suitable for numerous with target, the zone is complicated, each interregional border is not obvious, saturation degree brightness is hanged down is the remote sensing images of characteristics; Existing research work can only be handled the adjustment of two kinds of tones, and the effect of adjustment is still undesirable; And can not unify adjustment to the tone of the image that contains multiple color tones.

The present invention proposes a kind of unified adjustment algorithm of new masstone remote sensing images, can remove on a large scale the mixed and disorderly tangible amalgamation of color harmony border in the remote sensing images effectively, interactive operation is easy, and processing speed is fast, and can be applied to the other field of Flame Image Process.

A kind of new color pass-algorithm that proposes among the present invention carries out the adjustment unification of two kinds of tones; Be in the process that the sampled point of waiting to adjust picture point and reference picture matees; Introduce probability segmentation result based on expectation maximization as important judgment criteria, obtained matching result preferably.

In a word, use the adjustment unification that the present invention can fast and effeciently carry out two kinds of tones.The present invention has solved traditional length consuming time of the image adjusting method based on manual work well, and workload is big, inefficient deficiency, and the present invention is significantly increased on the agility of the convenience of user interactions, calculating and matching result.

Description of drawings

Fig. 1 is based on the multi-level image of color transmission and unifies the method for adjustment schematic flow sheet;

Fig. 2 (a) is a hue regions to be adjusted;

Fig. 2 (b) is with reference to hue regions;

Fig. 3 (a) is a K average segmentation result of waiting to adjust hue regions, and various colors is represented different cut zone;

Fig. 3 (b) is the K average segmentation result with reference to hue regions, and various colors is represented different cut zone;

Fig. 4 is the sample graph with reference to hue regions, the redness point expression sampled point among the figure;

Fig. 5 (a) treats the adjustment hue regions to carry out probability and cut apart synoptic diagram, when probability is cut apart, is split into the same area and representes with white in the drawings;

Fig. 5 (b) is cut apart synoptic diagram to carrying out probability with reference to hue regions, when probability is cut apart, is split into the same area and representes with white in the drawings;

Fig. 6 carries out the adjusted figure as a result of tone with the present invention.

Embodiment

Unifying method of adjustment based on the multi-level image of color transmission may further comprise the steps:

The piece image that 1) will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space;

2) with the K mean algorithm two width of cloth images are cut apart, and each cut zone of reference picture is sampled;

3) with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust between each cut section of image and set up mapping relations;

4) treat the convergent-divergent that the adjustment image carries out brightness, it and reference brightness are kept consistency;

5) treat each point of adjustment image, in the sampled point of reference picture, seek and its point that matees most, the color value of match point in the reference picture is passed to the corresponding point of waiting to adjust in the image;

Fig. 1 has showed the flow process of unifying method of adjustment based on the multi-level image of color transmission.

The described piece image that will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space, the steps include:

(1) piece image that has two kinds of tones is divided into two width of cloth images, is to exist the picture of two kinds of tones to carry out preliminary area dividing by the user to one, it is divided into two parts: picture to be adjusted (Fig. 2 (a)) and reference picture (Fig. 2 (b));

(2) if former picture uses is the RGB color space, then pass through following step, be converted into l α β color space, the steps include:

A) first step is to change into the XYZ tristimulus values to image from RGB, and wherein XYZ is the device independent space, and is as follows:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{ccc}0.5141& 0.3239& 0.1604\\ 0.2651& 0.6702& 0.0641\\ 0.0241& 0.1228& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

B) second step was to be converted into the LMS space from XYZ space, and is as follows:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3897& 0.6980& -0.0787\\ -0.2298& 1.1834& 0.0464\\ 0.0000& 0.0000& 1.0000\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]$

C) merge these two transformed matrixs, obtain transition matrix, obtain following matrix from the RGB color space to the LMS color space:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3811& 0.5783& 0.0402\\ 0.1967& 0.7244& 0.0782\\ 0.0241& 0.1288& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

D) calculate the logarithm value of LMS value:

L＝logL，M＝logM，S＝logS

Adopt following formula to carry out final conversion:

$\left[\begin{array}{c}l\\ \mathrm{\α}\\ \mathrm{\β}\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{3}}& 0& 0\\ 0& \frac{1}{\sqrt{6}}& 0\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& -2\\ 1& -1& 0\end{array}\right]\left[\begin{array}{c}L\\ M\\ S\end{array}\right]$

Describedly two width of cloth images are cut apart, and each cut zone of reference picture are carried out sampling step with the K mean algorithm:

(3) with the K mean algorithm two width of cloth images are cut apart, its step does;

E) point that will import is divided into K initial set, adopts disposable or all can heuristic cutting apart;

F) calculate the central point or the barycenter of each set;

G) utilize formula:

D _m=min|x _m-u _i| ², x _mBe the point of input, u _iBarycenter for each set

Each point is distributed to apart from its nearest central point;

H) recomputate the barycenter of each set, and utilize formula:

$V=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\underset{x_j\∈S_i}{\mathrm{\Σ}}{|x_j-u_i|}^2$

Calculate the difference of two squares in the cluster,, or after the step of iteration some, cut apart end when the value of its value less than certain user's appointment; Otherwise carrying out g and h step of iteration.

Showed among Fig. 3 treat the adjustment hue regions and cut apart with the K average with reference to hue regions after the result, the different colours among the figure is being represented different cut zone.

(4) each cut zone of reference picture is sampled, can sample by the user is manual, or the computer random sampling.Sampling to reference to hue regions is as shown in Figure 4, redness point expression sampled point wherein.

Describedly with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust and set up the mapping relations step between each cut section of image:

(5) use expectation-maximization algorithm that two width of cloth images are carried out the probability segmentation procedure respectively to be:

I) initialization: utilize the K average segmentation result of a last joint to calculate each mean value of areas u _iAnd standard deviation sigma _i

J) expectation: (x y) belongs to i Gauss model G to pixel I _i(i; u _i, σ _i) probability can pass through computes:

$i^P\mathrm{xy}=\frac{\exp (-\frac{{(I(x,y)-u_i)}^2}{2{\mathrm{\σ}}_i^2})}{{\mathrm{\Σ}}_{j=1}^N\exp (-\frac{{(I(x,y)-u_j)}^2}{2{\mathrm{\σ}}_j^2})}$

G _iRepresent regional i Gaussian distribution G _i(i; u _i, σ _i), _ip _Xy(x y) belongs to G to the pixel I of representative image I _iProbability, p _XyThe pixel I of representative image I (x, probability distribution estimation y), $p_{\mathrm{Xy}}^t=\left\{p_{\mathrm{Xy}}^t{}_i\right|i=1,2...\}.$

K) maximization: the Gaussian distribution G that recomputates regional i according to following formula _i(i; u _i, σ _i) average u _iAnd standard deviation sigma _i:

$u_i=\frac{1}{\underset{x,y}{{\mathrm{\Σ}}_i}{p}_{\mathrm{xy}}}\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}I(x,y)$

${\mathrm{\σ}}_i=\sqrt{\frac{\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}{(I(x,y)-u_i)}^2}{\underset{x,y}{{\mathrm{\Σ}}_i}{P}_{\mathrm{xy}}}}$

1) repeats j, two steps of k up to convergence.

Fig. 5 has showed that treating the adjustment color harmony carries out the synoptic diagram as a result after probability is cut apart with reference to hue regions, and the part that is split into the same area is filled with white in the drawings;

(6) to reference picture with wait to adjust between each cut section of image and set up mapping relations, step is:

M) we will set up the mapping function f () of each Gauss model

of waiting to adjust in the image some Gauss model in the reference picture; This mapping can be artificial the appointment; Also can be to accomplish automatically, but will satisfy the condition in the step 2 by algorithm;

N) suppose that image to be adjusted has two Gauss models With

In luminance channel $u_{t_i}\≥u_{t_j},$ So, the average u ' after the mapping also should satisfy ${u_{t_i}}^{\′}\≥{u_{t_j}}^{\′}.$ In our algorithm, when With

The brightness average the most approaching, and satisfy under the situation of monotonicity, Be mapped to

The described adjustment image convergent-divergent that carries out brightness of treating makes it and the reference brightness step that keeps consistency:

(7) because the difference of two width of cloth images in overall brightness if directly adopt their brightness to compare, will cause very poor matching result, therefore, we need in advance the brightness of image to be changed, and they are consistent.We have adopted a kind of linear mapping to come the average and the standard deviation of the Luminance Distribution of two width of cloth images are shone upon:

$Y\′\left(p\right)=\frac{{\mathrm{\σ}}_s}{{\mathrm{\σ}}_t}(Y\left(p\right)-u_t)+u_s$

u _s, σ _s: reference brightness average and standard deviation

u _t, σ _T:: brightness average and the standard deviation of waiting to adjust image

Described each point of treating the adjustment image is sought in the sampled point of reference picture and its point that matees most, and the color value of match point in the reference picture is passed to the corresponding point step of waiting to adjust in the image:

(8) treat each point of adjusting image, in the sampled point of reference picture, seek and its point that matees most, the steps include:

O) for corresponding sampled point in each point of waiting to adjust image and the reference picture, at first whether judgement point and the sampled point in the reference picture waiting to adjust in the image belongs to the same area after the K average is cut apart, and promptly judges f (KMeansPartitionResult (P _t)) whether equal f (KMeansPartitionResult (P _s)).If two points are not at the same area, promptly above two functional values are unequal, then continue a pair of point relatively down; If at the same area, promptly above two functional values equate, then continue step 2 and mate;

P) with the brightness of point-to-point transmission and probability distribution as evaluation criterion, calculate P _tAnd P _sDistance, P _tAnd P _sDistance function definition as follows:

$\mathrm{dis}\tan \mathrm{ce}(p_s,p_t)=w_1|l_{p_s}-l_{p_t}|+w_2\left|\right|{\stackrel{\→}{P}}_{p_s}-{\stackrel{\→}{P}}_{p_t}\left|\right|$

Expression p _s, p _tLuminance difference

Some p after expression is shone upon through the zone _tProbability segmentation result vector

represents

and

Euclidean distance between

Usually get w ₂∝ 10w ₁

Calculating waits to adjust distance between point and all sampled points in the reference picture in the image, and wherein that minimum sampled point of distance is optimal match point.

(9) color value with match point in the reference picture passes to the corresponding point of waiting to adjust in the image, the steps include:

Q) finding P _tMatch point P _sAfter, with P _sColouring information (being the value of α β passage) pass to P _t, keep P _tBrightness constant (being the value of l passage).

Last tone adjustment result is as shown in Figure 6, can find out, and two zones that tone is different of script, through unify the method for adjustment adjustment based on the multi-level image of color transmission after, tone reaches unanimity.

What more than enumerate only is specific embodiment of the present invention.Obviously, the invention is not restricted to above embodiment, many distortion can also be arranged.All distortion that those of ordinary skill in the art can directly derive or associate from content disclosed by the invention all should be thought protection scope of the present invention.

Claims (1)

1. the multi-level image based on the color transmission is unified method of adjustment, it is characterized in that may further comprise the steps:

The piece image that 1) will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space;

2) with the K mean algorithm two width of cloth images are cut apart, and each cut zone of reference picture is sampled;

3) with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust between each cut section of image and set up mapping relations;

4) treat the convergent-divergent that the adjustment image carries out brightness, it and reference brightness are kept consistency;

5) treat each point of adjustment image, in the sampled point of reference picture, seek and its point that matees most, the color value of match point in the reference picture is passed to the corresponding point of waiting to adjust in the image;

The described piece image that will have two kinds of tones is divided into two width of cloth images---reference picture and image to be adjusted, and with this two width of cloth image transitions to l α β color space step be:

(1) piece image that has two kinds of tones is divided into two width of cloth images, is to exist the picture of two kinds of tones to carry out preliminary area dividing by the user to one, it is divided into two parts: wait to adjust picture and reference picture;

(2) if former picture uses is the RGB color space, then pass through following step, be converted into l α β color space, the steps include:

A) first step is to change into the XYZ tristimulus values to image from RGB, and wherein XYZ is the device independent space, and is as follows:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{ccc}0.5141& 0.3239& 0.1604\\ 0.2651& 0.6702& 0.0641\\ 0.0241& 0.1228& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

B) second step was to be converted into the LMS space from XYZ space, and is as follows:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3897& 0.6980& -0.0787\\ -0.2298& 1.1834& 0.0464\\ 0.0000& 0.0000& 1.0000\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]$

C) merge these two transformed matrixs, obtain transition matrix, obtain following matrix from the RGB color space to the LMS color space:

$\left[\begin{array}{c}L\\ M\\ S\end{array}\right]=\left[\begin{array}{ccc}0.3811& 0.5783& 0.0402\\ 0.1967& 0.7244& 0.0782\\ 0.0241& 0.1288& 0.8444\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

D) calculate the logarithm value of LMS value:

L＝logL，M＝logM，S＝logS

Adopt following formula to carry out final conversion:

$\left[\begin{array}{c}l\\ \mathrm{\α}\\ \mathrm{\β}\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{3}}& 0& 0\\ 0& \frac{1}{\sqrt{6}}& 0\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& -2\\ 1& -1& 0\end{array}\right]\left[\begin{array}{c}L\\ M\\ S\end{array}\right];$

Describedly two width of cloth images are cut apart, and each cut zone of reference picture is carried out sampling step are with the K mean algorithm:

(3) with the K mean algorithm two width of cloth images are cut apart, its step does;

E) point that will import is divided into K initial set, adopts disposable or all can heuristic cutting apart;

F) calculate the central point or the barycenter of each set;

G) utilize formula:

D _m=min|x _m-u _i| ², x _mBe the point of input, u _iBarycenter for each set

Each point is distributed to apart from its nearest central point;

H) recomputate the barycenter of each set, and utilize formula:

$V=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\underset{x_j\∈S_i}{\mathrm{\Σ}}{|x_j-u_i|}^2$

Calculate the difference of two squares in the cluster,, or after the step of iteration some, cut apart end when the value of its value less than certain user's appointment; Otherwise carrying out g and h step of iteration;

(4) each cut zone of reference picture is sampled, can sample by the user is manual, or the computer random sampling;

Describedly with expectation-maximization algorithm two width of cloth images are carried out probability respectively and cut apart, and to reference picture with wait to adjust and set up the mapping relations step between each cut section of image and be:

(5) use expectation-maximization algorithm that two width of cloth images are carried out the probability segmentation procedure respectively to be:

I) initialization: utilize the K average segmentation result of a last joint to calculate each mean value of areas u _iAnd standard deviation sigma _i

J) expectation: (x y) belongs to i Gauss model G to pixel I _i(i; u _i, σ _i) probability can pass through computes:

$P_{\mathrm{xy}}^i=\frac{\exp (-\frac{{(I(x,y)-u_i)}^2}{2{\mathrm{\σ}}_i^2})}{{\mathrm{\Σ}}_{j=1}^N\exp (-\frac{{(I(x,y)-u_j)}^2}{2{\mathrm{\σ}}_j^2})}$

G _iRepresent regional i Gaussian distribution G _i(i; u _i, σ _i), _ip _Xy(x y) belongs to G to the pixel I of representative image I _iProbability, p _XyThe pixel I of representative image I (x, probability distribution estimation y),

K) maximization: the Gaussian distribution G that recomputates regional i according to following formula _i(i; u _i, σ _i) average u _iAnd standard deviation sigma _i:

$u_i=\frac{1}{\underset{x,y}{\mathrm{\Σ}}p_{\mathrm{xy}}^i}\underset{x,y}{\mathrm{\Σ}}P_{\mathrm{xy}}^{i}I(x,y)$

${\mathrm{\σ}}_i=\sqrt{\frac{\underset{x,y}{\mathrm{\Σ}}P_{\mathrm{xy}}^i{(I(x,y)-u_i)}^2}{\underset{x,y}{\mathrm{\Σ}}P_{\mathrm{xy}}^i}}$

L) repeat j, two steps of k are up to convergence;

(6) to reference picture with wait to adjust between each cut section of image and set up mapping relations, step is:

M) we will set up the mapping function f () of each Gauss model of waiting to adjust in the image some Gauss model in the reference picture; This mapping can be artificial the appointment; Also can be to accomplish automatically, but will satisfy the condition in the step 2 by algorithm;

N) so in luminance channel to suppose two Gauss models

and

are arranged image to be adjusted; Average u ' after the mapping also should satisfy

in our algorithm; Brightness average as

with

is the most approaching; And satisfy under the situation of monotonicity,

is mapped to

The described adjustment image convergent-divergent that carries out brightness of treating makes it and the reference brightness step that keeps consistency be:

(7) because the difference of two width of cloth images in overall brightness; If directly adopt their brightness to compare; Will cause very poor matching result, therefore, we need in advance the brightness of image to be changed; They are consistent, and we have adopted a kind of linear mapping to come the average and the standard deviation of the Luminance Distribution of two width of cloth images are shone upon:

$Y^{\′}\left(p\right)=\frac{{\mathrm{\σ}}_s}{{\mathrm{\σ}}_t}(Y\left(p\right)-u_t)+u_s$

u _s, σ _s: reference brightness average and standard deviation

u _t, σ _t: brightness average and the standard deviation of waiting to adjust image;

Described each point of treating the adjustment image is sought in the sampled point of reference picture and its point that matees most, and the color value of match point in the reference picture is passed to the corresponding point step of waiting to adjust in the image:

(8) treat each point of adjusting image, in the sampled point of reference picture, seek and its point that matees most, the steps include:

O) for corresponding sampled point in each point of waiting to adjust image and the reference picture, at first whether judgement point and the sampled point in the reference picture waiting to adjust in the image belongs to the same area after the K average is cut apart, and promptly judges f (KMeansPartitionResult (P _t)) whether equal f (KMeansPartitionResult (P _s)), if two points not at the same area, promptly above two functional values are unequal, then continue a pair of point relatively down; If at the same area, promptly above two functional values equate, then continue step 2 and mate;

P) with the brightness of point-to-point transmission and probability distribution as evaluation criterion, calculate P _tAnd P _sDistance, P _tAnd P _sDistance function definition as follows:

$\mathrm{dis}\tan \mathrm{ce}(p_s,p_t)=w_1|l_{p_s}-l_{p_t}|+w_2\left|\right|{\stackrel{\→}{P}}_{p_s}-{\stackrel{\→}{P}}_{p_t}\left|\right|$

Expression p _s, p _tLuminance difference

Expression point p _sProbability segmentation result vector

Some p after expression is shone upon through the zone _tProbability segmentation result vector

means

and

Euclidean distance between

Usually get w ₂∝ 10w ₁

Calculating waits to adjust distance between point and all sampled points in the reference picture in the image, and wherein that minimum sampled point of distance is optimal match point;

(9) color value with match point in the reference picture passes to the corresponding point of waiting to adjust in the image, the steps include:

Q) finding P _tMatch point P _sAfter, with P _sColouring information pass to P _t, keep P _tBrightness constant.

Images