CN101047869B - Method and device for correction gamma property of video communication - Google Patents

Abstract

This invention provides a correction method and a device for video communication gamma property, which obtains a brightness histogram of output brightness signals to convert it to distributed probability density function of the output brightness signals and determines the extremum point of the function and sets up a math relation between the extremum points of the density function of the distributed probability of the output bright signal brightness and the input one and utilizes the math relation to convert the math relation between the distributed probability density functions of the input and output signals and the Gamma property function to the math relation of output signal brightness distributed probability density function and the Gamma property function to resolve the converted relation to determine Gamma property parameter and correct the Gamma part.

Description

A kind of bearing calibration of gamma characteristic of video communication and device

Technical field

The present invention relates to the video communication technical field, be specifically related to a kind of bearing calibration and device of gamma characteristic of video communication.

Background technology

Video communication, especially multi-party video communication at present along with the developing rapidly of broadband network, obtains increasingly extensive application.At home and in the world, video conference and visual telephone service are becoming the basic service on the NGN (Next Generation Network, next generation network).The telecom operators of various countries also pay much attention to this market opportunity.Be expected in the coming years, video communication service will become the important business growth point of operator.

A key issue of development video communication business is: the user experience (User Experience or Quality of Experience) that improves end-to-end (End-to-end).In the user experience except parameters such as the QoS of network such as packet loss, delay, shake, the R factor, for video, because the distortion to luminance signal (Distortion) that Gamma (gamma) nonlinear problem that each link causes causes also is to influence the key factor that the end user experiences.

At present, mainly concentrate on assurance network QoS pre-process and post-process (Pre-processing, post-processing) aspect relevant for method that improves end-to-end user experience and technology with video compression coding.For the brightness distortion problem that the Gamma characteristic causes, lack the solution of concern and system.But the importance of this problem has caused the concern of some international big telecom operators.

Below the Gamma characteristic is briefly introduced.

The process of video communication is: it is first-class as video camera/shooting that the light signal of scene that need be transmitted such as personage, background, file etc. enters into video communication terminal (hereinafter referred terminal), convert data image signal to through A/D, pass through compressed encoding again, send out, arrive distant terminal, then, be reduced to data image signal through past compression (decompression) decoding, on display device, show again, finally become light signal again by the human eye perception.In said process, image brightness signal has passed through a plurality of links.According to definition, the input-output of the image brightness signal of a link of Gamma characteristic relation is not linear, but a kind of nonlinear relation.The image brightness signal here (Luminance) is a kind of luminance signal of broad sense, i.e. light signal at the beginning, to the signal of telecommunication, arrive digitized image brightness/grey scale signal again, the signal in each stage all contains the information of image brightness signal, therefore, in a broad sense, image brightness signal has passed through a plurality of links.

The universal model of single link Gamma characteristic as shown in Figure 1.

Among Fig. 1, the nonlinear relation of input luminance signal and output luminance signal can be expressed as: L _Out=G (L _In), wherein, L _OutBe output luminance signal, L _InBe input luminance signal, function G (.) is a nonlinear function.

Typical Gamma characteristic example as shown in Figure 2.

The numeral that marks in each square among Fig. 2 is a brightness value, and the gray scale of square is represented the bright degree of luminance signal.Among Fig. 2 (a), above the brightness of delegation's gray scale square be linear increment, promptly be incremented to 1.0 from 0.1, below the brightness of delegation's gray scale square increase progressively according to the power function rule, that is to say, below the brightness of delegation's gray scale square passed through the nonlinear distortion effect of Gamma.What provide among Fig. 2 (b) is Gamma characteristic with curve representation.

Get up when being together in series in other words when a plurality of link cascades (cascading), then total Gamma characteristic equals compound (composition) of each link Gamma function.

The universal model of the Gamma characteristic of a plurality of link cascades as shown in Figure 3.

Among Fig. 3, the nonlinear relation of the input luminance signal of each link and output luminance signal is respectively:

Lout＝G ⁽¹⁾(Lin)、Lout＝G ⁽²⁾(Lin)、Lout＝G ⁽³⁾(Lin)。

Can learn the compound of each link Gamma function thus for shown in the formula (1):

G _CT(.)＝G ⁽¹⁾(.)oG ⁽²⁾(.)oG ⁽³⁾(.)......G ^(n-1)(.)oG ⁽ⁿ⁾(.) (1)

l _out＝G _CT(l _in)＝G ⁽ⁿ⁾(G ^(n-1)(G ^(n-2)(.......G ⁽²⁾(G ⁽¹⁾(l _in)))))

Wherein, "." compound operation of representative function.CT represents cascaded total, i.e. the meaning of the total Gamma of cascade.

In practice, Gamma is non-linear is caused by different reasons.The Gamma characteristic of display device such as CRT monitor under ideal state is:

L _out＝L _in ^2.2 (2)

And the desirable Gamma characteristic of corresponding video camera/camera is:

L _out＝L _in ^0.45 (3)

From the origin of Gamma problem, the Gamma problem originates from CRT monitor, and is non-linear in order to compensate this because its Gamma value is 2.2, introduced Gamma value 0.45 in video camera.In this example, the form of Gamma characteristic is a power function (Power Function).Need to prove that the input and output luminance signal here all is to have carried out normalization (Normalized), i.e. 0≤L in coordinate space separately _Out≤ 1,0≤L _In≤ 1.And the display of other type is such as liquid crystal etc., the form of its Gamma function or can be different, though perhaps also be power function in form, the parameter difference.

Ideal situation is to have linear relationship between input luminance signal and the output luminance signal, i.e. L _Out=L _InObtain linear relationship, must carry out Gamma for link and proofread and correct (Gamma Correction) with non-linear Gamma characteristic.

To the Gamma correction principle figure of a Gamma link as shown in Figure 4.

Among Fig. 4, for a link, its Gamma characteristic is given to be L _Out=Gg (L _In), like this, can use another one correction link L _Out=Gc (L _In) and it carry out cascade, make total Gamma characteristic become real linear relationship, thereby reach the nonlinear purpose of proofreading and correct given link.

Obviously, Gg (.) and Gc (.) inverse function each other.In the ordinary course of things,, obtain its inverse function and not necessarily separate for a function, and, even there is inverse function, also can't obtain with Calculation Method.

In the practical application, more situation is the situation that has a plurality of Gamma links, to the Gamma correction principle figure of a plurality of Gamma links as shown in Figure 5.

Among Fig. 5, correction link need be inserted between former and later two given Gamma links.The Gamma characteristic of preceding given link is L _Out=Ga (L _In), the Gamma characteristic of the given link in back is L _Out=Gp (L _In).At this moment, the Gc (.) in the correction link is very complicated, no longer is that simple inverse function has concerned between Gc (.) and Ga (.) or the Gp (.).

Realize above-mentioned Gamma bearing calibration, its prerequisite is: can determine the Gamma characterisitic parameter for the cascade of a given Gamma link or a plurality of given Gamma link.The Gamma characterisitic parameter here is exactly the parameter of Gamma characterisitic function curve.

In communication generally speaking, correction need relate to plural communication terminal.Such as in one two side's video communication, the video of terminal A is sent to terminal B, and the correction of this road video just relates to Gamma link on the terminal A and the Gamma link on the terminal B simultaneously so.

At present, the method for determining the Gamma characterisitic parameter mainly contains two kinds:

Method one: apparatus measures method.Promptly measure some points on the Gamma characterisitic function curve, then, adopt the method for data fitting to carry out curve fitting, to determine the Gamma characterisitic parameter by instrumentation.

Method two: the method that adopts input luminance signal and output luminance signal.Promptly,, just can find Gg (.) is carried out the Gc (.) that Gamma proofreaies and correct as long as Gg (.) satisfies certain condition for single given Gamma link; For a plurality of given Gamma links,, just can find Ga (.) and Gp (.) are carried out the Gc (.) that Gamma proofreaies and correct as long as Ga (.), Gp (.) satisfy certain condition.

From the description of above-mentioned two kinds of methods as can be seen, the implementation method of determining the Gamma characterisitic parameter at present all has a precondition, promptly clearly know the input luminance signal of the Gamma link that need be determined the Gamma characterisitic parameter and the concrete numerical value of output luminance signal, just clearly know the input luminance signal of Gamma link and the A to Z of of output luminance signal, therefore, the implementation method of above-mentioned two kinds of definite Gamma characterisitic parameters all belongs to non-Blind Test metering method.

The realization principle of non-Blind Test metering method as shown in Figure 6.

Among Fig. 6, Gamma characterisitic parameter measuring system utilizes input luminance signal the A to Z of of Gamma link, output luminance signal the A to Z of to measure the Gamma characterisitic parameter, the Gamma link here can be single Gamma link, also can be the Gamma link of cascade.

But the non-Blind Test metering method scope of application in actual applications is very limited, that is to say, above-mentioned precondition is unappeasable under many circumstances, exemplifies the three kinds of common application that can not satisfy above-mentioned precondition at present below.

Application scenario one: for IPTV streaming media service and application such as (Internet Protocol Television), because in the program making process, be subjected to the influence of the Gamma characteristic of video input apparatus, when program broadcasts, especially situation such as program request, being used to gather the Gamma characteristic of the video input apparatus of vision signal when can't obtain original program making, and, also can not measure the Gamma characterisitic parameter of video input apparatus.

Application scenario two: also have the problems referred to above for application such as data conferencings.At present, the development of video conference and the development of data conferencing are synchronous, the combination that both are perfect, and (collaborativeapplications) has very big meaning for collaboration applications.In environment such as enterprise, above-mentioned collaboration applications business has the strong market demand.But, in data conferencing was used, a lot of multimedia documents can not be examined such as the source of picture etc., and very difficult acquisition generated the Gamma characteristic of the video input apparatus of these data at that time, and, also can not measure the Gamma characterisitic parameter of video input apparatus.

Application scenario three: for public video communication business, in order to reduce cost and video communication service use threshold often a large amount of employings cheap camera, especially those dog-cheap USB interface cameras towards ten million domestic consumer.The Gamma characteristic curve of the video input apparatus that these are cheap and the L of standard _Out=L _In ^0.45Fall far short, even be not the form of power function.And from the technical data that dispatches from the factory of these cheap cameras, generally can't obtain its Gamma characterisitic parameter.Even the basic technical data that just do not dispatch from the factory of some cheap camera.When the user uses these cameras at home, also can not obtain the Gamma characterisitic parameter by the method for above-mentioned definite Gamma characterisitic parameter.

More than three kinds of application be very important, and above-mentioned three kinds of application all have very big market potential, especially the development in IPTV and collaboration data meeting market is very fast.Video communication will really be used for huge market, must rely on the road of walking public operation to attract huge numbers of families, so just requires the condition of crossing the threshold must be very low, and the price of video input apparatus is very cheap.The non-Blind Test metering method of determining at present the Gamma characterisitic parameter is proofreaied and correct Gamma to be difficult to use, and the method for measuring by instrumentation has improved the realization cost of video communication service.

Summary of the invention

The objective of the invention is to, a kind of bearing calibration and device of gamma characteristic of video communication is provided, utilize the histogram of output luminance signal to determine the gamma characteristic parameter,, improve the purpose that Gamma proofreaies and correct ease for use to realize reducing the video communication cost.

For achieving the above object, the bearing calibration of a kind of gamma characteristic of video communication provided by the invention comprises:

A, obtain output luminance signal brightness histogram;

B, the brightness histogram that will export luminance signal are converted to output luminance signal Luminance Distribution probability density function, and determine the extreme point of output luminance signal Luminance Distribution probability density function;

Mathematical relationship between the extreme point of c, the extreme point of setting up output luminance signal Luminance Distribution probability density function and input luminance signal Luminance Distribution probability density function;

D, utilize between described extreme point and the extreme point mathematical relationship with input luminance signal, output luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function be converted to: at the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function and the Gamma characterisitic function;

E, the mathematical relationship after the described conversion is found the solution, to determine the gamma characteristic parameter;

F, the gamma link is carried out Gamma correction according to described gamma characteristic parameter.

Described gamma link is: the cascade combination of single given gamma link or a plurality of given gamma links.

Described step b comprises:

The brightness histogram of output luminance signal is converted to the output luminance signal Luminance Distribution probability density function of polynomial form:

f _r(r)=c _dr ^d+ c _D-1r ^D-1+ c _D-2r ^D-2+ ... ..+c ₁R+c ₀Wherein: c _d, c _D-1, c _D-2... .., c ₀Be d+1 multinomial coefficient, r is the amplitude of output luminance signal.

Described steps d comprises:

Input, output luminance signal Luminance Distribution probability density function f separately _e(x, t), f _r(x, t) and the mathematical relationship between the Gamma characterisitic function be:

D (e; P) f _r(r)=f _e(e); Wherein: r=g (e, p),

E is the amplitude of input luminance signal;

The extreme point quantity of input luminance signal, output luminance signal Luminance Distribution probability density function separately is identical, and quantitative mathematical relationship is: r _k=g (e _kP), p=[p ₁, p ₂, p ₃... .., p _M] ^T, p is a parameter vector, described parameter vector is made up of M parameter;

Wherein: k=1,2,3 ..., J; J is the extreme point quantity of Luminance Distribution probability density function;

At the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function after the conversion and derived function thereof and Gamma characterisitic function and inverse function thereof, the derived function is:

$z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J;$

Wherein: k=1,2,3 ..., J, J are the extreme point quantity of Luminance Distribution probability density function, r _kBe the extreme point of output luminance signal Luminance Distribution probability function, and, derived function

For: $\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}={\mathrm{dc}}_d{r}^{d-1}+(d-1)c_{d-1}{r}^{d-2}+(d-2)c_{d-2}{r}^{d-3}+.....+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="H06182861420060622E000021.png,H06182861420060622E000062.png"\; state="\{\{state\}\}">c</figure-callout>}}_1.$

Described step e comprises:

Utilize direct solving equation method or nonlinear function optimization method to mathematical relationship: $z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J$ Find the solution, to determine the gamma characteristic parameter.

The nonlinear function optimization method comprises:

Construct cost function according to the mathematical relationship between described output luminance signal Luminance Distribution probability density function and the Gamma characterisitic function:

$J\left(p\right)=\underset{k=1}{\overset{J}{\mathrm{\&Sigma;}}}{(z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}})}^2;$

In M parameter, choose M-1 parameter arbitrarily, the dimension of parameter space is reduced to the M-1 dimension;

The method of determining M-1 parameter is: determine parameter vector p _True, make for any p ∈ R ^M-1, concern J (p)＞=J (p _Ture) set up, promptly search out the overall smallest point of cost function, parameter vector p _TrueBe for this M-1 the determined numerical value of parameter;

Utilize and concern g (1; P)=1 in conjunction with p _True, determine surplus next gamma characteristic parameter.

Determine p _TrueMethod comprise: combined optimization method, neural net method, rough power searching method.

Described rough power searching method comprises step:

The gamma characteristic parameter space that dimension is reduced one dimension is divided into a plurality of hypercubes;

Choose initial search point, and begin to carry out traversal search from the hypercube at this initial search point place according to predefined procedure;

Calculate the geometric center coordinate Q of each hypercube that enters in the search procedure, and calculate J (Q) according to the hypercube that enters in the expression formula of described cost function and the search procedure;

If J (Q) is smaller or equal to predetermined threshold, the hypercube of perhaps searching for satisfies predetermined condition, and then the geometric center coordinate Q of the hypercube that this search is entered is as p _True, search procedure finishes;

Otherwise, continue search procedure.

Described initial search point is provided with according to the empirical value of gamma characteristic parameter in the practical application.

Described traversal search comprises: with the hypercube at initial search point place as initial hypercube, the hypercube that surrounds initial hypercube is divided into the multilayer hypercube array of the preceding one deck of parcel successively according to the distance apart from initial hypercube, and successively searches for.

In rough power searching method, the gamma characteristic parameter space of dimension reduction one dimension is divided into the hypercube of a plurality of coarsenesses, and described traversal search comprises:

, the hypercube that surrounds initial hypercube is divided into the multilayer hypercube array of one deck before the parcel successively according to the distance of the initial hypercube of distance, and successively searches for as initial hypercube with the hypercube at initial search point place;

The hypercube that satisfies condition that searches as new gamma characteristic parameter space, and is divided into more fine-grained hypercube with it, and the like, carry out search successively from coarse to fine.

The present invention also provides a kind of means for correcting of gamma characteristic, and described device comprises: obtain histogram module, first modular converter, extreme point computing module, memory module, second modular converter, Gamma characterisitic parameter and find the solution module and Gamma correction module;

Obtain the histogram module: be used to obtain the brightness histogram of output luminance signal, and export first modular converter to;

First modular converter: be used for the brightness histogram of the output luminance signal of its reception is converted to output luminance signal Luminance Distribution probability density function, and export the extreme point computing module and second modular converter to;

Extreme point computing module: be used to calculate each extreme point of the output luminance signal Luminance Distribution probability density function of its reception, and export second modular converter to;

Memory module: mathematical relationship and the storage input between the extreme point that is used to store output luminance signal Luminance Distribution probability density function and the extreme point of input luminance signal Luminance Distribution probability density function, export luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function;

Second modular converter: the mathematical relationship that is used for the extreme point stored according to its extreme point that receives, output luminance signal Luminance Distribution probability density function, memory module will import, export luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function be converted to: at the extreme point place, export the mathematical relationship between luminance signal Luminance Distribution probability density function and the Gamma characterisitic function;

The Gamma characterisitic parameter is found the solution module: be used for the mathematical relationship after the described conversion is found the solution calculating, to determine the gamma characteristic parameter;

Gamma correction module: be used for the gamma link being carried out Gamma correction according to described gamma characteristic parameter.

Described device is arranged in the video data source device and/or is arranged in the intermediate equipment of video communication network and/or is positioned at the video data destination device.The present invention also provides a kind of means for correcting of gamma characteristic, comprising:

Description by technique scheme as can be known, the histogram that Gamma bearing calibration provided by the invention only needs to export luminance signal gets final product, like this, Gamma parameter determination method of the present invention can be called total blindness's method of measurement; Because the present invention does not need any knowledge of input luminance signal, therefore, Gamma bearing calibration of the present invention has very high application feasibility; Gamma revision method of the present invention is specially adapted to IPTV, collaboration data meeting, is extensive use of the public video communication of low side video input apparatus; The present invention has adopted successively methods such as searching for, search for from coarse to fine, choose according to actual conditions initial search point when adopting rough power searching method to determine the gamma characteristic function parameters, improved search efficiency; Thereby realized that by technical scheme provided by the invention raising Gamma proofreaies and correct ease for use, widened the range of application that Gamma proofreaies and correct, improved the purpose of user experience and service quality.

Description of drawings

Fig. 1 is the model schematic diagram of link Gamma characteristic;

Fig. 2 (a) is a Gamma characteristic schematic diagram one;

Fig. 2 (b) is a Gamma characteristic schematic diagram two;

Fig. 3 is the model schematic diagram of the Gamma characteristic of a plurality of link cascades;

Fig. 4 is the Gamma correction principle schematic diagram to a Gamma link;

Fig. 5 is the Gamma correction principle schematic diagram to a plurality of Gamma links;

Fig. 6 is the realization principle schematic of non-Blind Test metering method of the prior art;

Fig. 7 is the realization principle schematic that the total blindness Gamma characterisitic parameter of the embodiment of the invention is determined method;

Fig. 8 is the brightness histogram exemplary plot of vision signal;

Fig. 9 is the schematic diagram that concerns between the input signal, output luminance signal Luminance Distribution probability density function of the embodiment of the invention;

Figure 10 (a) is a frame input picture;

Figure 10 (b) is a frame output image;

Figure 10 (c) is the corresponding relation schematic diagram between input signal, the output luminance signal Luminance Distribution probability density function extreme point;

Figure 11 is the output luminance signal Luminance Distribution probability density function schematic diagram that interpolation and data fitting obtain that passes through of the embodiment of the invention;

Figure 12 is the principle schematic of the rough power searching method of the embodiment of the invention;

The initial hypercube of Figure 13 embodiment of the invention and the schematic diagram of its peripheral multilayer hypercube under two-dimensional case;

Figure 14 is the rough power searching method schematic diagram of the embodiment of the invention.

Embodiment

In a lot of application scenarioss, the concrete numerical value of output luminance signal is that the A to Z of is as can be known, and the concrete numerical value of input luminance signal is unknowable, even any knowledge of input luminance signal is all unknowable.The present invention only utilized the knowledge of output luminance signal to determine the Gamma characterisitic parameter, to carry out the Gamma correction.Because in technical scheme of the present invention, do not utilize any knowledge of input luminance signal, so the method for determining the Gamma characterisitic parameter in the technical solution of the present invention can be called total blindness Gamma characterisitic parameter and determine method.

The realization principle that total blindness Gamma characterisitic parameter of the present invention is determined method as shown in Figure 7.

Among Fig. 7, determine the link of Gamma characteristic for needs, the present invention determines the Gamma characterisitic parameter of this link according to the knowledge of known output luminance signal.Determine in the method that at total blindness Gamma characterisitic parameter of the present invention the A to Z of of output luminance signal is known, still, the present invention might not utilize the A to Z of of output luminance signal.

After having determined the Gamma characterisitic parameter, just can carry out Gamma to the Gamma link and proofread and correct according to the Gamma characterisitic parameter according to the knowledge of output luminance signal.In the present invention, need to determine that the link of Gamma characteristic can be single given Gamma link, also can be the cascade combination of a plurality of given Gamma links.

Below in conjunction with accompanying drawing technical scheme provided by the invention is described in detail.

At first, the present invention need obtain the brightness histogram of output luminance signal, and brightness histogram as shown in Figure 8.Among Fig. 8, the brightness of setting image is 0 to 255 grade, and then different brightness degrees are all corresponding to a Luminance Distribution probability.

Histogram is the technical term of technical field of image processing, and in fact, histogram is exactly a kind of distribution probability density function of discrete form.

Vision signal is made up of the consecutive image of a frame one frame, and the histogram of output luminance signal can obtain from a certain two field picture.The histogrammic method that obtains luminance signal from image belongs to routine techniques, is not described in detail at this.

The histogram that obtains the output luminance signal also can carry out in other stage, as in the output luminance signal also in one-dimensional signal, obtain the histogram of output luminance signal, at this moment, the output luminance signal does not change into image.

Exist confidential relation between histogram information and the continuous distribution probability density function.In general, can directly obtain brightness histogram by continuous distribution probability density function; Conversely, by brightness histogram, also can obtain continuous distribution probability density function by methods such as data interpolating or matches.In fact, there are strict proportional numbers magnitude relation in histogram information and continuous distribution probability density function.Be described as follows with co-relation.

For the incompatible theory of cascaded series of single given Gamma link or a plurality of given Gamma links, all set of luminance signal be s (t) | t ∈ R, 0≤s (t)≤1}, wherein, R represents the set of all real numbers.That is to say that all set of luminance signal are all signal amplitudes (amplitude) smaller or equal to the set of 1 nonnegative value time signal.The value of the luminance signal here is non-negative, and the value of luminance signal is determined according to physical significance, because negative brightness does not have physical significance.Any signal is bound to satisfy signal amplitude smaller or equal to 1 condition after handling through normalization.The luminance signal here is the luminance signal on the universal significance, and therefore the description below all is suitable for for input luminance signal, output luminance signal.In order to describe for simplicity, be example with the output luminance signal below, the relation between histogram information and the continuous distribution probability density function is described.

Owing to have random disturbances, so these output luminance signals can be regarded random process as.The statistical property of these output luminance signals may have nothing in common with each other, and still, according to the statistical property of signal, particularly according to the distribution probability characteristic, can classify to output signal.Any signal all has a distribution probability density function corresponding with it as a random process, if random process is (here steadily be proper steadily) stably, this distribution probability density function and time are irrelevant so; If random process is not stably, this distribution probability density function may be relevant with the time so.Therefore, in general,, can use f for a random process s (t) (t ∈ R, 0≤s (t)≤1) _s(x, t), t ∈ R represents its distribution probability density function.If proper random process stably, then f _s(x, t), t ∈ R and t are irrelevant, and promptly the distribution probability density function does not change in time and changes, at this moment, f _s(x, t)=f _s(x).

The normalization processing method of signal is as follows:

The t ∈ R if a signal s (t) does not satisfy condition, 0≤s (t)≤1 so, need carry out normalization to this signal and handle, and makes it satisfy t ∈ R, 0≤s (t)≤1.For example: if the span of signal reality be [0, S _Max], the signal s after then normalization is handled _n(t) be:

s _n(t)＝s(t)/S _max (4)

Subscript n in the formula (4) is represented English normalized, looks like to be normalization.

Correspondingly, if signal is reverted to actual value from the value of normalization, promptly signal is carried out contrary normalization and handle, its computing formula is as follows:

s(t)＝S _maxs _n(t) (5)

According to the definition of distribution probability density function, the distribution probability density function has following attribute:

${\&Integral;}_{-\&infin;}^{+\&infin;}{f}_s(x,t)\mathrm{dx}=1,$ For any t

And (6)

f _s(x, t) 〉=0, for any t

And,, satisfy smaller or equal to 1 nonnegative value signal for signal amplitude:

f _s(x, t)=0, x＜0 or x＞1 (7)

That is to say that signal value is greater than 1 or be impossible less than 0, probability is zero.

As a natural deduction be exactly:

${\&Integral;}_0^1{f}_s(x,t)\mathrm{dx}=1,$ For any t (8)

According to the definition of probability density function, for very little siding-to-siding block length δ and interval [0,1] last 1 x ₀, f _s(x ₀, t) δ ≈ Prob{x ₀≤ s (t)≤x ₀+ δ } (9)

Perhaps equivalently

$f_s(x_0,t)\mathrm{\&delta;}\&ap;\mathrm{Prob}\{x_0-\frac{1}{2}\mathrm{\&delta;}\&le;s\left(t\right)\&le;x_0+\frac{1}{2}\mathrm{\&delta;}\}---\left(10\right)$

Wherein: symbol Prob represents probability (Probability).

Its meaning directly perceived is to say, at moment t, luminance signal drops on interval [x ₀, x ₀+ δ] or

Probability be approximately equal to f _s(x ₀, t) δ.This is a kind of method that the continuous distribution probability density function is become discrete probability density in fact.Hence one can see that, by continuous probability density function, can obtain the brightness histogram of signal by such discretization.

For the luminance signal of normalization, can be divided into N subinterval to [0,1] interval, the length in each subinterval is 1/N.K (k=0,1,2 ...., N-1) individual subinterval is [k/N, (k+1)/N].If N is enough big, 1/N is enough little, so, can think:

$\frac{1}{N}{f}_s(\frac{2k+1}{2N},t)=\mathrm{Prob}\{\frac{k}{N}\&le;s\left(t\right)\&le;\frac{k+1}{N}\},k=\mathrm{0,1,2},\&CenterDot;\&CenterDot;\&CenterDot;.,N-1---\left(11\right)$

So, can form a probability sequence (sequence):

$\left\{\frac{1}{N}{f}_s(\frac{2k+1}{2N},t)\right|k=\mathrm{0,1,2}...,N-1\}---\left(12\right)$

If signal reverts in the signal space of its non-normalization, as in video communication usually luminance signal get the integer of 0-255, totally 256 grades of brightness certainly, also can be generalized to 2 with luminance signal ^DThe situation of level brightness, at this moment, need be with the unit interval [0,1] Linear Mapping become set 0,1,2,3 ..., 2 ^D-2,2 ^D-1}, the corresponding expansion 2 in each subinterval ^DDoubly, become (1/N) 2 ^DSo corresponding probability sequence becomes continuous probability density function:

$\left\{h\left(k\right)\right|h\left(k\right)=\frac{1}{N}{f}_s(\frac{2k+1}{2N},t),k=\mathrm{0,1,2}...,N-1\}---\left(13\right)$

According to formula (8) and formula (10), obviously can draw:

$\underset{i=0}{\overset{2^D-1}{\mathrm{\&Sigma;}}}h\left(i\right)=1---\left(14\right)$

This probability sequence in the formula (13) just is called the histogram of luminance signal s (t).

Can obviously find out from above-mentioned derivation: histogram is directly to be obtained by the continuous distribution probability density function of luminance signals, conversely, the continuous distribution probability density function of luminance signal also can be by obtaining after the processing such as histogram process data interpolating, match.

Provide a histogrammic object lesson below.When the brightness of signal comprised 256 intensity levels, the corresponding relation of concrete numerical value was in probability sequence and the histogram:

h(0)＝0

h(1)＝0

….

h(64)＝0.005

h(65)＝0.006

…..

h(190)＝0.006

h(191)＝0.005

h(192)＝0.001

h(193)＝0

…

h(255)＝0

In the present invention, set the concrete manifestation form of gal sign indicating number characterisitic function and clearly know, can be from the present invention as selecting a kind of using two kinds of forms of embodiment.Can certainly be the function of other form, if satisfy continuously smooth and at least second order can lead.

Below with function y=g (x; P), p=[p ₁, p ₂..., p _M] ^TThe Gamma characteristic of representing the Gamma link of Gamma characterisitic parameter the unknown, the Gamma link here comprise the situation of the cascade combination of single Gamma link or a plurality of Gamma links.In the expression mode of above-mentioned Gamma characterisitic function, p=[p ₁, p ₂..., p _M] ^TBe a parameter vector, generally speaking, parameter vector is made up of M parameter.The all or part of of these parameters is to need to determine.Therefore, according to this very general form, the Gamma characterisitic function almost can be any type of function, as long as satisfy function is condition for continuous, and, in general, the Gamma characterisitic function is smooth leading, at least be that piecewise smooth can be led, therefore, suppose that the Gamma characterisitic function is rational about single order and the existence of second order derived function of variable x.The single order derived function of Gamma characterisitic function can be used following symbolic representation:

$d(x;p)=\frac{\mathrm{dg}(x;p)}{\mathrm{dx}}---\left(15\right)$

And, the Gamma characterisitic function, also should satisfy:

g(1；p)＝1 (16)

In general, the Gamma characterisitic function can be represented with following two kinds of modes commonly used:

Mode one, power function:

$y=g(x;p)=p_1{\mathrm{<figure-callout\; id="2"\; label="x\; p"\; filenames="H06182861420060622E000021.png,H06182861420060622E000053.png"\; state="\{\{state\}\}">x</figure-callout>}}^{p_{}}+{\mathrm{<figure-callout\; id="3"\; label="p"\; filenames="H06182861420060622E000021.png"\; state="\{\{state\}\}">p</figure-callout>}}_3,p={[{\mathrm{<figure-callout\; id="1"\; label="p"\; filenames="H06182861420060622E000021.png,H06182861420060622E000062.png"\; state="\{\{state\}\}">p</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="p"\; filenames="H06182861420060622E000021.png,H06182861420060622E000053.png"\; state="\{\{state\}\}">p</figure-callout>}}_2,p_3]}^T---\left(17\right)$

Mode two, polynomial function:

Y=g (x; P)=p ₁x ^K+ p ₂x ^K-1+ ... .+p _KX+p _K+1, wherein, p=[p ₁, p ₂, p ₃..., p _K+1] ^T(18)

Above-mentioned formula (18) also can be transformed into the expression-form of formula (19):

y＝g(x；p)＝p ₁(x-x ₀) ^K+p ₂(x-x ₀) ^K-1+....+p _K(x-x ₀)+p _K+1 (19)

Wherein, p=[p ₁, p ₂, p ₃..., p _K+1, x ₀] ^T

If represent input luminance signal and output luminance signal respectively with e (t) and r (t), so, each self-corresponding distribution probability density function of e (t) and r (t) is: f _e(x, t) and f _r(x, t), and, there is following relation: r (t)=g (e (t) between e (t), r (t) and the Gamma characterisitic function; P), p=[p ₁, p ₂, p ₃..., p _M] ^T

According to probability theory, can derive:

D (e; P) f _r(r)=f _e(e), wherein (20)

r＝g(e；p)， $\&ForAll;e\&Element;\left[\mathrm{0,1}\right]$

The detailed process of deriving formula (20) can be not described in detail in the present embodiment referring to probability books commonly used.

From formula (20) as can be seen, formula (20) and time variable t are irrelevant.In fact, Gamma characterisitic function itself and time variable are irrelevant, therefore, as long as in one period considerable time, measure a Gamma characterisitic parameter, just can use this group Gamma characterisitic parameter in whole communication process, as in IPTV, the Gamma characterisitic parameter of a program can be thought identical always, it is just passable to measure a Gamma characterisitic parameter at first at each program like this, at most.

After having obtained output luminance signal Luminance Distribution probability density function, need to calculate each extreme point of output luminance signal Luminance Distribution probability density function.Each extreme point of computing function belongs to routine techniques, is not described in detail in an embodiment of the present invention.

An important foundation of the present invention is: have very strict one-to-one relationship between the extreme point of the extreme point of input luminance signal Luminance Distribution probability density function and output luminance signal Luminance Distribution probability density function.This corresponding relation as shown in Figure 10.

In Figure 10, figure (a) is a frame of incoming video signal, figure (b) figure is a frame of the corresponding outputting video signal of figure (a), and two curves among the figure (c) are respectively input luminance signal Luminance Distribution probability density function and output luminance signal Luminance Distribution probability density function.

From figure (c) as can be seen, the extreme point of two curves be strictness one to one.This corresponding relation is:

e ₁-＞r ₁(maximal point), e ₂-＞r ₂(minimal point), e ₃-＞r ₃(maximal point), e ₄-＞r ₄(minimal point), e ₅-＞r ₅(maximal point).

Be without loss of generality, setting input luminance signal Luminance Distribution probability density function and exporting luminance signal Luminance Distribution probability density function all has J extreme point, is respectively e ₁, e ₂...., e _JAnd r ₁, r ₂.. ...., r _J

On the basis of this extreme point corresponding relation, the present invention is to the extreme point e of input luminance signal Luminance Distribution probability density function _k, and the extreme point r of corresponding output luminance signal Luminance Distribution probability density function _k(k=1,2 ..., J) carried out a rational approximate processing, promptly set formula (21) and set up:

r _k＝g(e _k；p)，p＝[p ₁，p ₂，p ₃，...，p _M] ^T k＝1，2，3，…，J (21)

Wherein: k=1,2,3 ..., J; J is the extreme point quantity of Luminance Distribution probability density function.

Because the extreme point quantity of output luminance signal Luminance Distribution probability density function is identical with the extreme point quantity of input luminance signal Luminance Distribution probability density function, so, the J here can think the extreme point quantity of output luminance signal Luminance Distribution probability density function, also can be the extreme point quantity of input luminance signal Luminance Distribution probability density function.

Strictly speaking, concern r _k=g (e _kP), p=[p ₁, p ₂, p ₃..., p _M] ^TBe invalid, this can be from the enterprising line justification of mathematics.But, in actual applications,, can find r through analysis for a large amount of actual video signals _kAnd g (e _kP) be very approximately equalised, therefore, the formula (21) of thinking that the present invention can be similar to is set up.Certainly, formula (21) also can change a little, increases an additivity constant term or scale factor etc. in high spirits as one side of equation.In the following description, be to be that formula (21) is that example is described with the mathematical relationship between the extreme point of the extreme point of output luminance signal Luminance Distribution probability density function and input luminance signal Luminance Distribution probability density function, after formula (21) changed a little, process that the gal code parameters is determined and following description are basic identical, are not described in detail at this.

From foregoing description as can be seen, there are following two relations in the extreme point of input luminance signal Luminance Distribution probability density function with the extreme point of output luminance signal Luminance Distribution probability density function: concern 1, geometry topological relations qualitatively, promptly there is one-to-one relationship between the extreme point of two Luminance Distribution probability density functions, that is to say that the quantity of the extreme point of two Luminance Distribution probability density functions is identical.Concern 2, quantitative mathematical relationship: r _k=g (e _kP), p=[p ₁, p ₂, p ₃..., p _M] ^TCertainly, the mathematical relationship here allows conversion a little.

Set f _e(x) and f _r(x) can lead continuously, and d (x; P) can lead continuously, i.e. g (x; P) Second Order Continuous can be led.Like this, the present invention carries out differentiate to formula (20) both sides, can obtain:

$z(e;p)f_r\left(r\right)+d^2(e;p)\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}=\frac{{\mathrm{df}}_e\left(e\right)}{\mathrm{de}},$ Wherein

r＝g(e；p)， (22)

$z(e;p)=\frac{d\left(d(e;p)\right)}{\mathrm{de}}\&ForAll;e\&Element;\left[\mathrm{0,1}\right]$

Because e ₁, e ₂... .., e _JBe f _e(x) extreme point, therefore, f on these extreme points _e(x) derivative is zero.Like this, there is following formula:

$\frac{{\mathrm{df}}_e\left(e\right)}{\mathrm{de}}{|}_{e=e_k}=0,k=\mathrm{1,2,3},...,J---\left(23\right)$

Be tied to form upright just like ShiShimonoseki in conjunction with formula (22) and formula (23):

$z(e_k;p)f_r\left(r\right)+d^2(e_k;p)\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}=0---\left(24\right)$

r＝g(e _k；p)，k＝1，2，3，....，J

Set up because formula (21) is approximate, so formula (24) can be transformed to following relation:

$z(e_k;p)f_r\left(r_k\right)+d^2(e_k;p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J---\left(25\right)$

$z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J---\left(26\right)$

In the formula (26), k=1,2,3 ..., J, J are the extreme point quantity of Luminance Distribution probability density function, r _kBe the extreme point of output luminance signal Luminance Distribution probability function, and, derived function

For: $\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}={\mathrm{dc}}_d{r}^{d-1}+(d-1)c_{d-1}{r}^{d-2}+(d-2)c_{d-2}{r}^{d-3}+.....+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="H06182861420060622E000021.png,H06182861420060622E000062.png"\; state="\{\{state\}\}">c</figure-callout>}}_1.$

In actual applications, the Gamma characterisitic function all is dull, and therefore, there is inverse function in the Gamma characterisitic function.This inverse function can be remembered and makes g ^-1(x; P), obviously, the inverse function of Gamma characterisitic function also depends on gamma parameter vector p.

As can be seen, do not contain any information relevant for input signal in the formula (26) from formula (26), formula (26) depends on the extreme point of output luminance signal Luminance Distribution probability density function and output luminance signal Luminance Distribution probability density function fully.Thereby the present invention can only determine the Gamma characterisitic parameter according to output signal and Luminance Distribution probability density function thereof.

The histogram of setting the output luminance signal of the present invention's acquisition is: { h _r(k) | k=0,1,2 ...., N-1}.That is to say that include N in each histogram, in the histogram term, each is called " post " (bin).

Can obtain output luminance signal Luminance Distribution probability density function by the output image brightness histogram according to formula (12):

$\left\{h_r\left(k\right)\right|h_r\left(k\right)=\frac{1}{N}{f}_r\left(\frac{2k+1}{2N}\right),k=\mathrm{0,1,2}...,N-1\},$ So:

$f_r\left(\frac{2k+1}{2N}\right)={\mathrm{Nh}}_r\left(k\right),k=\mathrm{0,1,2}...,N-1---\left(27\right)$

For N the point that is evenly distributed on [0,1] interval $a_k=\frac{2k-1}{2N},k=\mathrm{1,2},....,N,$ Can obtain function f _r(x) numerical value on these discrete points, promptly $f_r\left(\frac{2k+1}{2N}\right)k=\mathrm{0,1,2}...,N-1.$ The position of these discrete points on coordinate system r axle as shown in figure 11.

If N is enough big, so, can obtain exporting the expression formula of Luminance Distribution probability density by interpolation or data fitting mode:

f _r(r)＝c _dr ^d+c _d-1r ^d-1+c _d-2r ^d-2+.....+c ₁r+c ₀ (28)

Wherein: c _d, c _D-1, c _D-2... .., c ₀Be d+1 multinomial coefficient, r is the amplitude of output luminance signal.

This expression formula is the multinomial that comprises polynomial splines.

In the ordinary course of things, for the image of 256 grades of brightness, N=256, at this moment, N is enough big.The implementation method of interpolation or Technology of Data Fitting has multiple, and the present invention does not limit in the specific implementation scope of interpolation or data fitting.

The derivative of formula (28) is:

$\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}={\mathrm{dc}}_d{r}^{d-1}+(d-1)c_{d-1}{r}^{d-2}+(d-2)c_{d-2}{r}^{d-3}+.....+c_1---\left(29\right)$

The present invention can select the expression formula of the gamma characteristic function of any form of expression that provides in (17), (18), (19) for use, at this moment, has obtained the concrete form of Gamma characterisitic function, and just parameter vector p need determine.Calculate in the formula (26) any one for each specified value of parameter vector p, promptly calculate each multiplier and plus item mutually in the formula (26).

If J equation arranged, as long as J 〉=M-1, wherein, M is for needing to determine the number of gamma characteristic parameter, so, necessarily can solve unique separating by this set of equations as the determined value of M-1 gamma parameter wherein, then, utilize formula (16) to determine constant term again, promptly determine a remaining gamma parameter as a constraints.

That is to say, because the existence of formula (16), make and all satisfy a constraints between all gal code parameters, these gal code parameters are if M, so, having only M-1 parameter is independently, as long as determine any M-1 parameter in M the gal code parameters, a remaining gamma parameter is found the solution by formula (16) and is got final product, thereby has determined all parameters of Gamma characterisitic function.

In the ordinary course of things, condition J 〉=M-1 can both satisfy.

The method that obtains parameter vector p according to formula (26) has multiple, mainly introduces the method for two kinds of definite parameter vector p below:

Method one, direct solving equation method.

Obtain J equation by formula (26), therefrom select M-1 equation arbitrarily, form the equation group simultaneous solution.In general, this set of equations is non-linear, and surmounts (Transcendental), as for Gamma characterisitic function of power function form etc.Therefore there are not analytic solutions (closed-formsolution).Need with numerical solution (numerical solution method).Numerical solution about equation group belongs to routine techniques, is not describing in detail in the present embodiment.

Method two, nonlinear function optimization method.

Construct a cost function according to formula (26):

$J\left(p\right)=\underset{k=1}{\overset{J}{\mathrm{\&Sigma;}}}{(z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}})}^2---\left(30\right)$

Obviously, for the real parameter vector p of Gamma characterisitic function _True, should make J (p in theory _True)=0 is because parameter vector p _TrueSatisfy each equation in the formula (30), therefore, each sum term in the formula (30) all is zero.But in actual applications, because exist sum of errors approximate as obtain approximate etc. in the histogram process, so, each sum term in the formula (30) can all not equal zero, but, should be a very little numerical value, and should satisfy following condition: for any p ∈ R ^M, J (p) 〉=J (p is all arranged _True).That is to say p _TrueIt is the overall smallest point (global minimal point) of function J (p).

Can know that from foregoing description the Gamma characterisitic function for any one form that adopts formula (17), (18), (19) to provide can calculate g according to each given numerical value of parameter vector p ^-1(r _kP), z (g ^-1(r _kP); P), d ²(g ^-1(r _kP); P), simultaneously, output luminance signal Luminance Distribution probability density function is known, like this, just can calculate f according to formula (28), (29) _r(r _k) and Thereby just can calculate each sum term in the formula (30), also just can calculate total summed result J.

Therefore, in the nonlinear function optimization method, determine that the problem of parameter vector p just changes into the mathematical problem of cost function J (p) being asked its overall smallest point.

In the nonlinear function optimization method, can adopt following three kinds of modes to determine parameter vector p.

Mode (1), conventional mathematical optimization method.

Owing to have derivative among the J (p), therefore, can adopt classical mathematical optimization method such as gradient method, conjugate gradient method to wait to determine parameter vector p.

Determine that by conventional mathematical optimization method the detailed process of parameter vector p belongs to routine techniques, is not described in detail in the present embodiment.

Mode (2), neural net method.

Determine that by neural net method the detailed process of parameter vector p belongs to routine techniques, is not described in detail in the present embodiment.

Mode (3), rough power searching method (Brutal Force Search Method).So-called rough power search as its name suggests, is exactly all possibilities of exhaustive search.

The situation of quantizing for parameter, then parameter might value set be finite aggregate, so, each point in the search set one by one just can find to make and the point of J (p) minimum like this, just found overall smallest point p _TrueBut, this situation seldom, in most cases, parameter is got successive value, therefore, parameter might value set be unlimited set, can't really carry out exhaustive search.

Get the situation of successive value for parameter, rough power searching method is that parameter space is divided into a plurality of little hypercubes (Hypercube), then, in each hypercube, get a point as sampled point, as the geometric center point of hypercube etc., last, the functional value of calculation cost function on the sampled point of each hypercube, find to make the sampled point of hypercube of cost function minimum, with this sampled point as overall smallest point.

Below at parameter get the situation of successive value, in conjunction with the accompanying drawings to utilizing rough power searching method to determine that the implementation procedure of the parameter of Gamma characterisitic function is described in detail.

Though rough power searching method is " stupid " a bit,, rough power searching method has extensive use in engineering practice, and such as code breaking etc., effective method still is rough power searching method.And for the mathematical optimization problem, all there is the problem that is absorbed in local minimum (local minimalpoint) in existing technology except rough power search, but there is not this problem in rough power search method.

In the present invention, can utilize priori, find suitable initial search point, like this, can reduce the number of times that needs search greatly, thereby improve the efficient of rough power searching method about parameter.

Rough power searcher ratio juris as shown in Figure 12.

Among Figure 12, M-1 parameter might value set constituted parameter space (ParameterSpace is called for short PS), parameter space is M-1 dimension theorem in Euclid space R ^M-1A subclass.After having determined parameter space, the implementation method of rough power search comprises the steps:

Step 1, hypercube are divided.

PS is divided into a plurality of M-1 hypercubes (Hypercube), in the accompanying drawing 12, ABCDEFGH, 8 points are formed a hypercube.Because each parameter range varies in size, so each length of side of hypercube is also different.

Set k (k=1,2 ..., M-1) individual parameter p _kSpan be [min _k, max _k], k dimension carried out P _kFive equilibrium, thus the length of side of each hypercube is on this dimension

${\mathrm{\&Delta;}}_k=\frac{{\max }_k-{\min }_k}{P_k}---\left(31\right)$

Thereby the volume of each hypercube is:

$V={\mathrm{\&Pi;}}_{k=1}^{M-1}{\mathrm{\&Delta;}}_k={\mathrm{\&Pi;}}_{k=1}^{M-1}\left(\frac{{\max }_k-{\min }_k}{P_k}\right)---\left(32\right)$

Therefore, total hypercube number greater than $T={\mathrm{\&Pi;}}_{k=1}^{M-1}{P}_k.$ This numerical value is possible maximum, if the shape of PS is not a hypercube, its long cube number that comprises may be far smaller than so $T={\mathrm{\&Pi;}}_{k=1}^{M-1}{P}_k.$

For each hypercube indicator vector I=[i ₁, i ₂...., i _M-1] ^TRepresent, wherein i _k(k=1,2 ..., M-1, i _k=1,2,3 ...., P _k) be illustrated on k the dimension, this hypercube is i _kIndividual.

Therefore, for I=[i ₁, i ₂...., i _M-1] ^TIndividual hypercube, its coordinate range on each dimension is:

[min _k+(i _k-1)Δ _k，min _k+i _kΔ _k] (33)

The geometric center Q of this hypercube _ICoordinate be:

[min ₁+(i ₁-1/2)Δ ₁，min ₂+(i ₂-1/2)Δ ₂，..............，min _M-1+(i _M-1-1/2)Δ _M-1] ^T (34)

Step 2, choose initial search point.

In general, for certain parameter p _k, all there is a more reasonably value, can be used as initial value.When adopting the power function form as the Gamma characterisitic function, for video input apparatus, the value of Gamma parameter is generally about 2.2, because it is 2.2 that industrial standard requires, reason owing to manufacturing technology and product quality, the Gamma parameter may positive and negatively depart from 2.2, but as a rule, relatively near 2.2.Like this, if begin search with 2.2 as initial search point, because 2.2 relatively near actual value, the number of times that then finds actual value to attempt is just fewer.Same reason can be each parameter p _k(k=1,2 ..., M-1) find its suitable initial value p _{Int k}, these initial values form a vector so, are exactly the initial value p of parameter vector p _Int=[p _{Int k}, p _{Int k}, p _{Int k}...., p _{Int (M-1)}] ^T

Step 3 begins search according to the initial search point of choosing.

At first judge p _IntDrop in which hypercube, can determine p by comparing methods such as coordinate _IntWhich drop in the hypercube.Relatively the specific implementation process of coordinate method is:

The coordinate of setting this hypercube is I _Int=[i _Int1, i _Int2...., i _{Int (M-1)}] ^T, Rule of judgment is: for k=0, and 1,2 ... ..M-1 the time, and if only if, and following formula (35) is set up, and then determines p _IntDropping on coordinate is: I _Int=[i _Int1, i _Int2...., i _{Int (M-1})] ^THypercube in.

min _k+(i _intk-1)Δ _k≤p _intk＜min _k+i _intkΔ _k (35)

Behind the hypercube of having determined the initial search point place, begin search from this hypercube.Calculate J (p according to formula (30) _Int), if J is (p _Int) formula (36) is set up, the hypercube of perhaps searching for satisfies predetermined condition, as the quantity of the hypercube of search for reaches predetermined value etc., and so, step 5 is arrived in whole search procedure end.At this moment, p _True=p _IntWherein, p _TrueBe global optimum's point, i.e. the parameter vector of the final Gamma characterisitic function that obtains.

J(p _int)≤J _threshold (36)

Wherein, J _ThresholdIt is a threshold value given in advance.

If J is (p _Int) formula (36) is set up, then arrive step 4.

Step 4, continuation search.

Continuation search in step 4 can be that layering is carried out.The hypercube that is enclosed in initial hypercube outside can be one or more layers.

Under two-dimensional case, initial hypercube and its peripheral multilayer hypercube are as shown in Figure 13.

Among Figure 13, the square of Intermediate grey is initial hypercube, with the cube of the limit adjacency of initial hypercube be the ground floor hypercube, with the cube of the limit adjacency of ground floor hypercube be second layer hypercube, with the cube of the limit cheapness of second layer hypercube be the 3rd layer of hypercube.

Hierarchical search method of the present invention is: search for each hypercube in each layer outside the initial hypercube one by one, in the search of each layer hypercube, should travel through each hypercube in this layer according to predefined procedure.Predefined procedure can be diversified, and the present invention does not limit the form of predefined procedure, as long as the hypercube that can travel through in one deck is just passable.

In certain one deck hypercube search procedure, when searching certain hypercube according to predefined procedure, need calculate the coordinate of the geometric center Q of this hypercube according to formula (34), then, computing function value J (Q) is if J (Q) can make formula (37) set up, the quantity of the hypercube of perhaps searching for reaches predetermined value, so, whole search procedure finishes, to step 5.At this moment, p _True=Q;

J(Q)≤J _threshold (37)

If J (Q) can not make formula (37) set up, and the quantity of the hypercube of searching for does not reach predetermined value yet, continue in this layer hypercube, to search for according to predefined procedure.If the search of this layer hypercube is finished and the J (Q) of each hypercube of this layer all can not make formula (37) set up, then continue search one deck hypercube down.

After having searched for the L layer, no matter whether find the geometric center Q of the hypercube of satisfy condition (37), search procedure all will finish, at this moment, and should be with the Q among the J (Q) of the minimum that searches as p _TrueTo step 5.The L here is a threshold value given in advance, and expression needs the number of plies of search at most.

Step 5, search procedure finishes.

Above-mentioned searching method can also be applied to reach the fastest best search effect according in the search procedure from coarse to fine.

Search procedure from coarse to fine as shown in Figure 14.

Among Figure 14, at first, carry out the search first time to step 5 according to above-mentioned steps one.Search for the first time can be regarded coarse search as, and like this, what the length of side of the hypercube in the parameter space can be provided with is bigger, like this, and the negligible amounts of hypercube in the parameter space.In search procedure from coarse to fine, whether the predetermined condition that the hypercube of searching in the step 3 satisfies can be coarser than the predetermined division granularity for granularity of division.If the hypercube of satisfy condition (36) or (37) has been found in search for the first time, then search procedure finishes.Search procedure can adopt the method for hierarchical search for the first time.

If do not search the hypercube of satisfy condition (36) or (37) in the search procedure in the first time, and, the granularity division of hypercube does not also reach prescribed particle size, then in the corresponding hypercube of the J (Q) of the minimum that searches for the first time, carry out for the second time thinner search, at this moment, the hypercube that the J (Q) of the minimum that should be searched the first time is corresponding is as new entire parameter space.The length of side of each hypercube in the new parameter space of this moment has diminished, and repeats the search procedure of above-mentioned steps one to step 5.Equally, in for the second time thinner search procedure, if found the hypercube of satisfy condition (36) or (37), search procedure finishes for the second time.Search procedure can adopt the method for hierarchical search for the second time.

If in for the second time thinner search procedure, do not find the hypercube of satisfy condition (36) or (37), and, the granularity division of hypercube does not also reach prescribed particle size, then in the corresponding hypercube of the J (Q) of the minimum that searches for the second time, carry out thinner for the third time search, the rest may be inferred, and the geometric center of the hypercube of satisfy condition (36) or (37) that last fine search is found is put p as global optimum _True, i.e. the parameter vector of Gamma characterisitic function.

If the granularity division of hypercube has reached prescribed particle size, still still do not find the hypercube of satisfy condition (36) or (37), then search procedure finishes.At this moment, p is put as global optimum in the set center of hypercube that should the J (Q) of the minimum that finds is corresponding _True, i.e. the parameter vector of Gamma characterisitic function.

In above-mentioned search procedure each time from coarse to fine, can adopt the method for hierarchical search.

Behind the parameter vector of having determined the Gamma characterisitic function by said method, just can carry out Gamma and proofread and correct the Gamma link.Here, the link that need carry out the Gamma correction can be video data source equipment, also can be the intermediate equipment in the video communication network, can also be the video data destination device.

From the description of technique scheme as can be seen, Gamma bearing calibration provided by the invention only needs to know the histogram of output luminance signal and the Gamma characterisitic parameter that one group of loose assumed condition just can be determined given Gamma link, thereby for multimedia communications system provides a kind of Gamma bearing calibration that is easy to realize, Gamma bearing calibration provided by the invention has very high application feasibility, thereby widened the range of application that Gamma proofreaies and correct greatly, especially can be at IPTV, the collaboration data meeting, the public video communication that is extensive use of the low side video input apparatus provides good Gamma calibration function, improve user experience and service quality greatly, further promote the competitiveness of above-mentioned business, be telecom operators, service provider and equipment vendors bring huge economic benefit.

The means for correcting of gamma characteristic of video communication provided by the invention mainly comprises: obtain histogram module, first modular converter, extreme point computing module, memory module, second modular converter, Gamma characterisitic parameter and find the solution module and Gamma correction module.

Obtain the histogram module and be mainly used in the brightness histogram that obtains the output luminance signal, and export the brightness histogram of its acquisition to first modular converter.Obtain the histogram module and can before the output luminance signal is converted to the output map picture frame, obtain the brightness histogram of output luminance signal, also can from the output map picture frame, obtain the brightness histogram of output luminance signal.Concrete as the description in the above-mentioned method.

The brightness histogram that first modular converter is mainly used in the output luminance signal of its reception is converted to output luminance signal Luminance Distribution probability density function, and will export luminance signal Luminance Distribution probability density function and export the extreme point computing module and second modular converter respectively to.The output luminance signal Luminance Distribution probability density function here can be f _r(r)=c _dr ^d+ c _D-1r ^D-1+ c _D-2r ^D-2+ ... ..+c ₁R+c ₀

The extreme point computing module is mainly used in after receiving output luminance signal Luminance Distribution probability density function, calculates each extreme point of output luminance signal Luminance Distribution probability density function, and each extreme point that will calculate acquisition transfers to second modular converter.The method of calculating the extreme point of output luminance signal Luminance Distribution probability density function belongs to routine techniques, is not described in detail at this.

Mathematical relationship between the extreme point that memory module is mainly used in storage output luminance signal Luminance Distribution probability density function and the extreme point of input luminance signal Luminance Distribution probability density function, and the storage input, export luminance signal separately the Luminance Distribution probability density function and the mathematical relationship d (e between the Gamma characterisitic function; P) f _r(r)=f _e(e).

Mathematical relationship between the extreme point of the output luminance signal Luminance Distribution probability density function here and the extreme point of input luminance signal Luminance Distribution probability density function can for:

r _k＝g(e _k；p)，p＝[p ₁，p ₂，p ₃，...，p _M] ^T。Certainly, the mathematical relationship here allows conversion a little, and is concrete as the description in the above-mentioned method.

Second modular converter be mainly used in according to the mathematical relationship of the extreme point of storing in its extreme point that receives, output luminance signal Luminance Distribution probability density function, the memory module with the input in the memory module, output luminance signal separately the Luminance Distribution probability density function and the mathematical relationship d (e between the Gamma characterisitic function; P) f _r(r)=f _e(e) be converted to: at the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function and derived function thereof and Gamma characterisitic function and inverse function thereof, single order derived function, the second order derived function: $z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J;$

Wherein: k=1,2,3 ..., J, J are the extreme point quantity of Luminance Distribution probability density function, r _kBe the extreme point of output luminance signal Luminance Distribution probability function, and, derived function

For: $\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}={\mathrm{dc}}_d{r}^{d-1}+(d-1)c_{d-1}{r}^{d-2}+(d-2)c_{d-2}{r}^{d-3}+.....+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="H06182861420060622E000021.png,H06182861420060622E000062.png"\; state="\{\{state\}\}">c</figure-callout>}}_1.$

The Gamma characterisitic parameter is found the solution module and is mainly used in the mathematical relationship after second modular converter conversion is found the solution calculating, to determine the gamma characteristic parameter.The module of finding the solution the Gamma characterisitic parameter can adopt direct solving equation method, nonlinear function optimization method to wait to determine the gamma characteristic parameter.The nonlinear function optimization method here comprises conventional mathematical optimization method, neural net method, rough power search etc.The Gamma characterisitic parameter is found the solution module when adopting rough power searching method, can adopt the hierarchical search method, also can adopt from coarse to fine hierarchical search method etc., the description in the concrete as above-mentioned method.

The Gamma correction module is mainly used in to be found the solution the Gamma characterisitic parameter that module obtains according to the Gamma characterisitic parameter gamma link is carried out Gamma correction.The gal sign indicating number link here comprises: the cascade combination of single given Gamma link or a plurality of given Gamma links.

Device provided by the invention is arranged in video equipment, as is arranged in the video data source device, is arranged in the intermediate equipment of video communication network, is arranged in the video data destination device for another example.

Though described the present invention by embodiment, those of ordinary skills know, the present invention has many distortion and variation and do not break away from spirit of the present invention, and the claim of application documents of the present invention comprises these distortion and variation.

Claims (11)

1. the bearing calibration of a gamma characteristic of video communication is characterized in that, comprising:

A, obtain output luminance signal brightness histogram;

B, the brightness histogram that will export luminance signal are converted to output luminance signal Luminance Distribution probability density function, and determine the extreme point of output luminance signal Luminance Distribution probability density function;

Mathematical relationship between the extreme point of c, the extreme point of setting up output luminance signal Luminance Distribution probability density function and input luminance signal Luminance Distribution probability density function;

D, utilize between described extreme point and the extreme point mathematical relationship with input luminance signal, output luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function be converted to: at the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function and the Gamma characterisitic function;

E, the mathematical relationship after the described conversion is found the solution, to determine the gamma characteristic parameter;

F, the gamma link is carried out Gamma correction according to described gamma characteristic parameter;

Described step b comprises:

The brightness histogram of output luminance signal is converted to the output luminance signal Luminance Distribution probability density function of polynomial form:

f _r(r)＝c _dr ^d+c _d-1r ^d-1+c _d-2r ^d-2+.....+c ₁r+c ₀；

Wherein: c _d, c _D-1, c _D-2... .., c ₀Be d+1 multinomial coefficient, r is the amplitude of output luminance signal;

Described steps d comprises:

Input luminance signal, output luminance signal Luminance Distribution probability density function f separately _e(x, t), f _r(x, t) and the mathematical relationship between the Gamma characterisitic function be:

d(e；p)f _r(r)＝f _e(e)；

Wherein: r=g (e, p),

E is the amplitude of input luminance signal;

The extreme point quantity of input luminance signal, output luminance signal Luminance Distribution probability density function separately is identical, and quantitative mathematical relationship is:

r _k=g (e _kP), p=[p ₁, p ₂, p ₃... .., p _M] ^T, p is a parameter vector, described parameter vector is made up of M parameter;

Wherein: k=1,2,3 ..., J; J is the extreme point quantity of Luminance Distribution probability density function;

At the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function after the conversion and derived function thereof and Gamma characterisitic function and inverse function thereof, the derived function is:

$z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}}=0,k=\mathrm{1,2,3},....,J;$

Wherein: k=1,2,3 ..., J, J are the extreme point quantity of Luminance Distribution probability density function, r _kBe the extreme point of output luminance signal Luminance Distribution probability function, and, derived function

For:

$\frac{{\mathrm{df}}_r\left(r\right)}{\mathrm{dr}}={\mathrm{dc}}_d{r}^{d-1}+(d-1)c_{d-1}{r}^{d-2}+(d-2)c_{d-2}{r}^{d-3}+.....+c_1.$

2. the method for claim 1 is characterized in that, described gamma link is: the cascade combination of single given gamma link or a plurality of given gamma links.

3. the method for claim 1 is characterized in that, described step e comprises:

Utilize direct solving equation method or nonlinear function optimization method to mathematical relationship:

Find the solution, to determine the gamma characteristic parameter.

4. method as claimed in claim 3 is characterized in that, the nonlinear function optimization method comprises:

Construct cost function according to the mathematical relationship between described output luminance signal Luminance Distribution probability density function and the Gamma characterisitic function:

$J\left(p\right)=\underset{k=1}{\overset{J}{\mathrm{\&Sigma;}}}{(z(g^{-1}(r_k;p);p)f_r\left(r_k\right)+d^2(g^{-1}(r_k;p);p)\frac{{\mathrm{df}}_r\left(r_k\right)}{\mathrm{dr}})}^2$

In M parameter, choose M-1 parameter arbitrarily, the dimension of parameter space is reduced to the M-1 dimension;

The method of determining M-1 parameter is: determine parameter vector p _True, make for any p ∈ R ^M-1, concern J (p)＞=J (p _Ture) set up, promptly search out the overall smallest point of cost function, parameter vector p _TrueBe for this M-1 the determined numerical value of parameter;

Utilize and concern g (1; P)=1 in conjunction with p _True, determine surplus next gamma characteristic parameter.

5. method as claimed in claim 4 is characterized in that, determines p _TrueMethod comprise: combined optimization method, neural net method, rough power searching method.

6. method as claimed in claim 5 is characterized in that, described rough power searching method comprises step:

The gamma characteristic parameter space that dimension is reduced one dimension is divided into a plurality of hypercubes;

Choose initial search point, and begin to carry out traversal search from the hypercube at this initial search point place according to predefined procedure;

Calculate the geometric center coordinate Q of each hypercube that enters in the search procedure, and calculate J (Q) according to the hypercube that enters in the expression formula of described cost function and the search procedure;

If J (Q) is smaller or equal to predetermined threshold, the hypercube of perhaps searching for satisfies predetermined condition, and then the geometric center coordinate Q of the hypercube that this search is entered is as p _True, search procedure finishes;

Otherwise, continue search procedure.

7. method as claimed in claim 6 is characterized in that, described initial search point is provided with according to the empirical value of gamma characteristic parameter in the practical application.

8. method as claimed in claim 6, it is characterized in that, described traversal search comprises: with the hypercube at initial search point place as initial hypercube, the hypercube that surrounds initial hypercube is divided into the multilayer hypercube array of the preceding one deck of parcel successively according to the distance apart from initial hypercube, and successively searches for.

9. method as claimed in claim 6 is characterized in that:

In rough power searching method, the gamma characteristic parameter space of dimension reduction one dimension is divided into the hypercube of a plurality of coarsenesses, and described traversal search comprises:

, the hypercube that surrounds initial hypercube is divided into the multilayer hypercube array of one deck before the parcel successively according to the distance of the initial hypercube of distance, and successively searches for as initial hypercube with the hypercube at initial search point place;

The hypercube that satisfies condition that searches as new gamma characteristic parameter space, and is divided into more fine-grained hypercube with it, and the like, carry out search successively from coarse to fine.

10. the means for correcting of a gamma characteristic of video communication, it is characterized in that described device comprises: obtain histogram module, first modular converter, extreme point computing module, memory module, second modular converter, Gamma characterisitic parameter and find the solution module and Gamma correction module;

Obtain the histogram module: be used to obtain the brightness histogram of output luminance signal, and export first modular converter to;

First modular converter: be used for the brightness histogram of the output luminance signal of its reception is converted to output luminance signal Luminance Distribution probability density function, and exporting the extreme point computing module and second modular converter to, this output luminance signal Luminance Distribution probability density function is: f _r(r)=c _dr ^d+ c _D-1r ^D-1+ c _D-2r ^D-2+ ... ..+c ₁R+c ₀, wherein: c _d, c _D-1, c _D-2... .., c ₀Be d+1 multinomial coefficient, r is the amplitude of output luminance signal;

Extreme point computing module: be used to calculate each extreme point of the output luminance signal Luminance Distribution probability density function of its reception, and export second modular converter to; Each extreme point of described output luminance signal Luminance Distribution probability density function is r _k=g (e _kP), p=[p ₁, p ₂, p ₃... .., p _M] ^T, p is a parameter vector, described parameter vector is made up of M parameter; K=1,2,3 ..., J, J are the extreme point quantity of Luminance Distribution probability density function;

Memory module: mathematical relationship between the extreme point that is used to store output luminance signal Luminance Distribution probability density function and the extreme point of input luminance signal Luminance Distribution probability density function and storage input luminance signal, output luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function, described input luminance signal, the extreme point quantity of exporting luminance signal Luminance Distribution probability density function separately are identical; Described input, output luminance signal Luminance Distribution probability density function f separately _e(x, t), f _r(x, t) and the mathematical relationship between the Gamma characterisitic function be:

d(e；p)f _r(r)＝f _e(e)；

Wherein: r=g (e, p),

E is the amplitude of input luminance signal;

Second modular converter: the mathematical relationship that is used for the extreme point stored according to its extreme point that receives, output luminance signal Luminance Distribution probability density function, memory module with input luminance signal, output luminance signal separately the Luminance Distribution probability density function and the mathematical relationship between the Gamma characterisitic function be converted to: at the extreme point place, the mathematical relationship between output luminance signal Luminance Distribution probability density function and the Gamma characterisitic function

Wherein: r _kBe the extreme point of output luminance signal Luminance Distribution probability function, derived function

For:

The Gamma characterisitic parameter is found the solution module: be used for the mathematical relationship after the described conversion is found the solution calculating, to determine the gamma characteristic parameter;

Gamma correction module: be used for the gamma link being carried out Gamma correction according to described gamma characteristic parameter.

11. device as claimed in claim 10 is characterized in that, described device is arranged in the video data source device and/or is arranged in the intermediate equipment of video communication network and/or is positioned at the video data destination device.

Images