CN1964502A - Data driven reconfigurable color image processing system - Google Patents

Abstract

The disclosed color image processing system comprises: a data memory for different coefficients, a multiplication/addition unit to multiply pixel color value with loaded coefficient and add results, a lookup unit for nonlinear mapping, a mode matching control unit to determine processing mode and download coefficients into last two units. This invention has wide application, such as DC, film, and security system.

Description

The color picture processing system that data-driven is reconfigurable

Technical field

The present invention relates to a kind of image processing system, particularly relate to a kind of treatment system of coloured image color of pixel value.

Background technology

Bayer chromatic filter array is the general form of modern coloured image digital collection.As shown in Figure 1, half of total pixel is green (G), and red (R) and blue (B) respectively accounts for 1/4th.

In order to obtain all colouring informations, cover color image sensor with the arrangement that repeats with redness, green or blue electric-wave filter.This arrangement or the sequence of filter can change, and still, that extensively adopts arranges in Kodak invention " Bayer ", and it is that 2 * 2 of repetition is arranged.

In order to show and other processing, video compression for example as shown in Figure 2, should be carried out following the processing stage, comprising: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling.Should be noted that the order that to put upside down between some stage, for example color interpolation and white balance.The physical treatment of carrying out in each stage outlines as follows:

1. defect correction: when cmos sensor or ccd sensor obtained the Bayer color format, the pixels different with the pixel response around it may be a kind of upsets, and normally unacceptable.This means that these pixels are defective, and if do not proofread and correct, what manifest after handling is the mistake of snowization shape.By use as the method for medium filtering or average value filtering to carrying out interpolation in their contiguous data that accurately write down, estimate these disappearances or defective pixels.

2. color interpolation: this processing mainly is to use the color value of necessary bearing calibration for two disappearances of each pixel insertion.Many algorithms of simulating and comparing are arranged, and it mainly comprises: bilinear interpolation, intermediate value interpolation, the bilinear interpolation of gradient calibration, Kodak rebuild substantially, edge sensing interpolation+and level and smooth tone conversion, Kodak edge strength algorithm, parameter gradient method, pattern identification method and have the interpolation algorithm of color correction.

3. white balance: people's vision system has the ability that white colour is mapped as the white sensation, even when shining with Different Light, object has different radiancy.In other words, if people holds the card of a white outside, then it looks like white.If he or she takes it under the indoor fluorescent lamp, it looks like white.Forward under the incandescent bulb, it is white that card seems to remain, even be put under the yellow bulb, it seems it also is white in a few minutes.In these light sources each down, the spectrum of the card of white reflection different colours, still, brain is enough clever, and to make that it looks like white.In transducer, this can handle by so-called white balance and realize that it comes down to change in the following manner the value of RGB:

R_w＝W_r ^*R_d

G_w＝W_g ^*G_d (1)

B_w＝W_b ^*B_d

Wherein W_r, W_g and W_b are respectively the coefficients of balance of conversion R_d, G_d and B_d.R_w, G_w and B_w handle the new rgb value that provides by white balance.The value of coefficient of balance W_r, W_g and W_b depends on lighting condition (temperature) and system environments, and should be adjustable.

4. color correction:, therefore, after white balance, need color correction process on all cmos image sensors because the response of the color filter array that uses is not the just in time response of the vision system of match people.Can obtain the value of the RGB in this stage by following matrix computations, that is:

$\left(\begin{array}{c}R\\ G\\ B\end{array}\right)=\left(\begin{array}{ccc}{C}_{\mathrm{rr}}& C_{\mathrm{rg}}& C_{\mathrm{rb}}\\ C_{\mathrm{gr}}& C_{\mathrm{gg}}& C_{\mathrm{gb}}\\ C_{\mathrm{br}}& C_{\mathrm{bg}}& C_{\mathrm{bb}}\end{array}\right)\left(\begin{array}{c}R\_w\\ G\_w\\ B\_w\end{array}\right)---\left(2\right)$

Wherein $C=\left(\begin{array}{ccc}{C}_{\mathrm{rr}}& C_{\mathrm{rg}}& C_{\mathrm{rb}}\\ C_{\mathrm{gr}}& C_{\mathrm{gg}}& C_{\mathrm{gb}}\\ C_{\mathrm{br}}& C_{\mathrm{bg}}& C_{\mathrm{bb}}\end{array}\right)$ Be the color correction coefficient matrix, and should determine according to different systems and condition.

5.Gamma adjust (brightness/contrast adjustment): total brightness and contrast of the processing controls image in this stage.The image that does not have suitably to proofread and correct seem be fade otherwise be dark.The amount that changes the Gamma correction not only can change brightness, and can change ratio red and green and blueness.Similar with white balance and color correction, output and the input of adjusting stage can be written as:

$\left(\begin{array}{c}{R}^{\′\′}\\ G^{\′\′}\\ B^{\′\′}\end{array}\right)=\left(\begin{array}{ccc}{D}_{\mathrm{rr}}& D_{\mathrm{rg}}& D_{\mathrm{rb}}\\ D_{\mathrm{gr}}& D_{\mathrm{gg}}& D_{\mathrm{gb}}\\ D_{\mathrm{br}}& D_{\mathrm{bg}}& D_{\mathrm{bb}}\end{array}\right)\left(\begin{array}{c}{R}^{\′}\\ G^{\′}\\ B^{\′}\end{array}\right)+\left(\begin{array}{c}{D}_r\\ D_g\\ D_b\end{array}\right)---\left(3\right)$

Wherein $\left(\begin{array}{ccc}{D}_{\mathrm{rr}}& D_{\mathrm{rg}}& D_{\mathrm{rb}}\\ D_{\mathrm{gr}}& D_{\mathrm{gg}}& D_{\mathrm{gb}}\\ D_{\mathrm{br}}& D_{\mathrm{bg}}& D_{\mathrm{bb}}\end{array}\right)$ With $\left(\begin{array}{c}{D}_r\\ D_g\\ D_b\end{array}\right)$ It is the corresponding parameters of adjusting.

6. the color conversion from RGB to YUV:, need convert yuv format to for further video compression for three kinds of color RGB of each pixel.This conversion realizes by following equation (4):

Y＝0.29900R+0.58700G+0.11400B

Cb＝-0.16874R-0.33126G+0.50000B+2 ^SP/2

Cr＝0.50000R-0.41869G-0.08131B+2 ^SP/2 (4)

" SP " representative sample precision wherein.Use above-mentioned equation, can obtain the preliminary equation of YCbCr to RGB, it can be used for decoding processing.

R＝Y+1.40200Cr

G＝Y-0.34414(Cb-2 ^SP/2)-0.71414(Cr-2 ^SP/2) (5)

B＝Y+1.72200(Cb-2 ^SP/2)

7. to the down-sampling of YUV: the visual impression according to processed image is carried out down-sampling.The ratio that is generally used for down-sampling is " 4: 2: 0 ", this means not carry out down-sampling on the Y component, and as shown in Figure 3 the Cb component is carried out the vertical and horizontal down-sampling of " 2: 1 " with the Cr component.

With the down-sampling ratio of selecting, the mode that has several realization down-samplings to handle.The easiest method is the value of duplicating a pixel simply, and skips other contiguous pixels.The advantage of this method is that the thing of complexity is oversimplified, and shortcoming is to produce serious edge effect.More complicated method is to use some filtering algorithms.A kind of good compromise proposal is to do average treatment between contiguous pixel.

Can see, all the processing stage relate to many parameters, and handling property mainly depends on the value of these parameters, these parameters change with different lighting conditions and sensing system again.The more important thing is that the parameter in different disposal stage can be offset each other.In other words, the optimal treatment in each stage can not guarantee optimal treatment integrally.In addition, if use DSP or asic chip to realize these processing, some the processing stage, for example color interpolation and defect correction, the complexity of complexity of calculation and programming is very big.The reduction of complexity is normally with the cost that is downgraded to of performance.Although developed the method for many these problems of processing,, expecting to have very much a kind ofly can provide better compromise system between complexity, performance, adaptability and robustness.

Summary of the invention

The objective of the invention is the problem that exists for the aspects such as complexity that overcome existing color picture processing system global optimization, calculating and programming when the processed pixels color value, a kind of better balanced color image processing system and processing method of having obtained between complexity of calculation, performance, adaptability and robustness is provided.In order to realize this purpose, the technical solution used in the present invention is:

Provide a kind of color picture processing system according to a first aspect of the present invention, comprising: data storage is used to store various coefficients required when carrying out different disposal; The multiplications/additions unit, be used for pending color of pixel value respectively with the corresponding multiplication that is written into from described data storage, and with the results added of being taken advantage of; Lookup unit is used for Nonlinear Mapping is carried out in the output of described multiplications/additions unit; The pattern configurations control unit is used for determining the tupe of system's execution, and when carrying out determined tupe coefficient is written into described multiplications/additions unit and described lookup unit from described data storage.

Described multiplications/additions unit and lookup unit can also be one-level, secondary, more than three grades or three grades.Preferably, system comprises secondary multiplications/additions unit and lookup unit, is connected and composed in turn by multiplications/additions unit, lookup unit, multiplications/additions unit and lookup unit; The quantity of each grade multiplication/adder unit and lookup unit is corresponding with the quantity of pending pixel; And each multiplications/additions unit is connected with the pattern configurations control unit with data storage respectively with lookup unit, and data storage is connected with the pattern configurations control unit simultaneously, to receive coefficient data under the control of pattern configurations control unit.From described pending color of pixel value multiply each other in first order multiplications/additions unit with the corresponding coefficient that from described data storage, is written into respectively and with the result of product addition after, carry out Nonlinear Mapping by first order lookup unit; The output of described first order lookup unit multiply each other once more in multiplications/additions unit, the second level with the corresponding coefficient that from data storage, is written into respectively and with the result of product addition after, carry out Nonlinear Mapping by second level lookup unit again, the result is the color of pixel value after the processing.

Perhaps preferably also comprise multiplexer and demodulation multiplexer; Each color of pixel value is transferred to the multiplications/additions unit behind multiplexer, through carry out multiplying with the corresponding coefficient that from described data storage, is written into and with the result of product addition after, carry out Nonlinear Mapping by lookup unit; Under the control that repeats instruction, repeating multiplication, addition and Nonlinear Mapping by described multiplications/additions unit and described lookup unit handles, after the processing of finishing all pixel color value, by the treated color of pixel value of demodulation multiplexer output.

Thereby described coefficient is to use the coefficient value of a plurality of adjacent pixels by the predetermined optimization of off-line training.Described multiplications/additions unit is made up of a plurality of multipliers and an accumulator, perhaps is made up of a multiplier and adder and carries out repeatedly multiplication and addition with repetitive mode.Described lookup unit can be a plurality of, or one and repeatedly operation; The content of the table in the described lookup unit and size are predetermined before using; Described Nonlinear Mapping can be the Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions.Described pattern configurations control unit is determined the tupe that system carries out according to user's instruction, and described tupe comprises full tupe, single tupe and multiprocessing pattern; In described full tupe, the input of system is original RGB data, and the data of the YUV420 form of system's output down-sampling; In described single tupe, system can be configured to carry out following color treatments any in the stage: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling; In described multiprocessing pattern, system can be configured to carry out the more than one color treatments stage.

Provide a kind of color image processing method according to a second aspect of the present invention, may further comprise the steps: obtain and store the required various coefficients of execution different disposal in advance by off-line training; Determine tupe according to user instruction; According to determined tupe its corresponding coefficient of pending pixel value is carried out multiplying, and with the results added of being taken advantage of; Result to addition carries out Nonlinear Mapping.The color value of wherein treating processed pixels is carried out the processing of multiplication, addition and Nonlinear Mapping, can be one-level, secondary, more than three grades or three grades.Wherein said tupe comprises full tupe, single tupe and multiprocessing pattern; In described full tupe, the input of system is original RGB data, and the data of the YUV420 form of system's output down-sampling; In described single tupe, system can be configured to carry out following color treatments any in the stage: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling; In described multiprocessing pattern, system can be configured to carry out the more than one color treatments stage.Described Nonlinear Mapping can be the Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions.Described can be by a plurality of multiplier executed in parallel to its corresponding coefficient execution of pending pixel value multiplying, also can be to be repeated by a multiplier.Described Nonlinear Mapping can also can be to be repeated by a lookup unit by a plurality of lookup unit executed in parallel.

Adopt the color picture processing system of technical solution of the present invention, and their ASIC or DSP realizes comparing with the existing color Processing Algorithm, and obtained beneficial effect is mainly reflected in following aspect:

(1) system of the present invention can be configured to by selecting tupe (full tupe, single tupe and multiprocessing pattern) to carry out different tasks simply, and need not any further programming and instruction, and this is absolute demand in any DSP realizes.On the other hand, the realization of system of the present invention can be designed as ASIC, can reach more performance (low-power and small size) in view of the above.This means that system of the present invention can realize providing better trading off about adaptability and performance than current ASIC and DSP;

(2) under different lighting conditions, do not need the user to adjust any parameter reaching more performance,, pre-determined all coefficient datas from the angle of total optimization because on the basis of having considered all illuminations and relevant environmental condition.This means that also under different conditions, system of the present invention is robust more;

(3) system of the present invention have a distribution owing to the integral body of all operations unit connects but the processing feature of model-free.In fact, both do not done any hypothesis, do not made any model yet for the relation between input color and the output color.Determine the input color uniquely and export all information that concern between the color by optimized coefficient data;

(4) can use the pixel of much more vicinity to obtain optimized coefficient data, so that provide more performance in off-line training step in online operational phase.This feature is very difficult in many traditional algorithms based on modeling, because determine that in parameter stage and actual operational phase all comprise the pixel of equal number.The use of more pixels means in the training of traditional scheme and operational phase has all increased complexity.Yet,, use more pixel only to cause the increase of training stage complexity, but the operational phase complexity do not influenced system of the present invention;

(5) system of the present invention is in architecture and application, its scale, size, size and number of unit can be amplified arbitrarily, be dwindled, because these key element unit (multiplications/additions unit and lookup unit) can add in the hardware or the simple recursion instruction by form of software repeats, so that system can handle more pixel simultaneously or more performance is provided.

These key features are arranged, and system of the present invention can provide new instrument, carries out better Color Image Processing in many application, for example digital camera, film, visual telephone and safety system.

Below in conjunction with accompanying drawing, color picture processing system of the present invention is done detailed explanation by specific embodiment.

Description of drawings

Fig. 1 is the Bayer chromatic filter array of coloured image digital collection general format;

Fig. 2 is the various processing stage of doing in order to show etc. and to coloured image;

Fig. 3 is a schematic diagram of Cb component and Cr component being carried out the vertical and horizontal down-sampling of " 2: 1 ";

Fig. 4 is the structural representation according to the color picture processing system of the embodiment of the invention;

Fig. 5 a and 5b are respectively the schematic diagrames according to two kinds of different structures of the multiplications/additions unit of embodiments of the invention;

Fig. 6 a and Fig. 6 b are respectively the figures according to the lookup unit structural representation of the embodiment of the invention and part Nonlinear Mapping function (1. linear function, 2. piecewise function and 3. sigmoid function);

Fig. 7 is the schematic diagram according to the pattern configurations control unit of the embodiment of the invention;

Fig. 8 is the another kind of structural representation according to the color picture processing system of the embodiment of the invention;

Fig. 9 is the flow chart of data processing figure according to the color picture processing system of the embodiment of the invention.

Among the figure: 401. data storages, 402. the multiplications/additions unit, 403. lookup unit, 404. pattern configurations control units, 412. multiplier, 422. accumulator, 432. adders, 413. data address generators, 423. storage device, 801. multiplexer, 802. repeat instruction, 803. demodulation multiplexers.

Embodiment

Embodiment 1: as shown in Figure 4, mainly be made up of four parts according to the color picture processing system of first embodiment of the invention: data storage 401, multiplications/additions unit 402, lookup unit 403 and pattern configurations control unit 404.

Data storage 401 is used for storage system when carrying out different disposal required various coefficients; Multiplications/additions unit 402 be used for pending color of pixel value respectively with the corresponding multiplication that is written into from described data storage 401, and with the results added of being taken advantage of; Lookup unit 403 is used for Nonlinear Mapping is carried out in the output of described multiplications/additions unit; Pattern configurations control unit 404 is used for determining the tupe of system's execution, and when carrying out determined tupe coefficient is written into described multiplications/additions unit 402 and described lookup unit 403 from described data storage 401.

The multiplications/additions unit of color picture processing system of the present invention and lookup unit can also be one-level, secondary, more than three grades or three grades.Preferably, comprise secondary multiplications/additions unit and lookup unit (just in the array that constitutes by multiplications/additions unit and lookup unit shown in Fig. 4 two row multiplications/additions unit and two row lookup unit being arranged) in the present embodiment, connect and compose in turn by multiplications/additions unit, lookup unit, multiplications/additions unit and lookup unit; The quantity of each grade multiplication/adder unit and lookup unit (line number of multiplications/additions unit and lookup unit array among Fig. 4 just) is corresponding with the quantity of pending pixel; And each multiplications/additions unit is connected with the pattern configurations control unit with data storage respectively with lookup unit, and data storage is connected with the pattern configurations control unit simultaneously, to receive coefficient data under the control of pattern configurations control unit.From pending pixel (X (i, j) to X (i+K, j+L)) color (for example RGB or YUV) value respectively with the corresponding coefficient that from described data storage 401, is written into first order multiplications/additions unit (the multiplications/additions unit of first row among Fig. 4 just) multiply each other in 402 and with the result of product addition after, carry out Nonlinear Mapping by first order lookup unit (lookup unit of secondary series among Fig. 4 just) 403; The output of the 402 pairs of first order lookup unit 403 in multiplications/additions unit, the second level (tertial multiplications/additions unit among Fig. 4 just) and the corresponding coefficient that is written into from data storage 401 carry out multiplying and once more with after the result of product addition, carry out Nonlinear Mapping by second level lookup unit (lookup unit of the 4th row among Fig. 4 just) 403 once more, the result is the color of pixel value after the processing.

The input of system be pending pixel (i, j) and its adjacent pixels (i+k, j+l) (k＜K, the color value of l＜L).System by carry out corresponding pattern configurations control unit 404 and operating unit provide pixel (s, t) and its adjacent pixels (s+m, t+n) (m≤M, the desired color value of n≤L).

Shown in Fig. 5 a, each multiplications/additions unit 402 is made up of a plurality of multipliers 412 and accumulator 422, perhaps shown in Fig. 5 b, form and under the control of repetitive instruction, carry out repeatedly multiplication and addition repeatedly, X among the figure in the mode that repeats respectively by a multiplier 412 and adder 432 _i(i=1,2 ..., N) be color component, C _i(i=1,2 ..., N) be predetermined coefficient.The input-output relation of multiplications/additions unit 402 is:

$Z_{\mathrm{out}}=\underset{i=1}{\overset{I}{\mathrm{\Σ}}}{C}_i*Z_{\mathrm{in}}\left(i\right)---\left(6\right)$

Wherein the number of times of multiplication, C are carried out in " I " representative in this unit _iBe the coefficient that from data storage 401, is written into, Z _In(i) be another factor of multiplication, it is from the color value of system's input or is the output of lookup unit 403.

Shown in Fig. 6 a, lookup unit 403 is made up of data address generator 413 and storage device 423, is input to output from it and carries out predetermined Nonlinear Mapping.In fact, the content of the table that all are relevant and size were scheduled to before using system of the present invention, for the data of input lookup unit 403, after data address generator 413 addressing, from the corresponding data of output output of lookup unit 403.On the basis of this specific character, a plurality of lookup unit can be replaced by a lookup unit, and repeatedly operation.Possible Nonlinear Mapping can be Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions, and Fig. 6 b shows the figure of linear function (1.), piecewise function (2.) and sigmoid function (3.).The characteristic of this nonlinear function also relates to determining of coefficient data.The more important thing is, in some mode of operation, can walk around lookup unit or make lookup unit invalid.In other words, lookup unit can be considered to have the gain unit of unit gain in these patterns.

As shown in Figure 7, pattern configurations control unit 404 determines to carry out which kind of tupe by system according to user's instruction.After determining tupe, the configuration remaining element, i.e. coefficient data memory 401, multiplications/additions unit 402 and lookup unit 403 are to carry out accordingly the processing to the pixel color data.System of the present invention has following operator scheme:

(1) full tupe

In this pattern, the input of system is original RGB data (for example Bayer form), and the YUV420 form of system's output down-sampling.

(2) single tupe

In this pattern, any during system can be configured to carry out following processing stage the: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling.The result is that 8 subpatterns are arranged in this single tupe.The input and output of these 8 subpatterns are different.For example: if select the color interpolation subpattern, system exports to the color value that each pixel provides two losses, and the input of system is a color value of the same pixel of proofreading and correct by the defect correction stage.Similarly, the result of this defect correction can obtain by another kind of system, and perhaps the system of the previous proposition by having the subpattern that corresponding defect correction handles obtains.

(3) multiprocessing pattern

In the multiprocessing pattern, system of the present invention can be configured to carry out the more than one color treatments stage.As an illustration, system can carry out defect correction and color interpolation simultaneously by selecting corresponding pattern simply and being written into corresponding coefficient from data storage 401.In this case, the input of system is original RGB data (color value of each pixel), and output is the RGB data (three color values of each pixel) of interpolation.And, also can carry out the combination of three phase process, promptly defect correction, color interpolation and white balance are handled.In this case, the input and output of system are respectively original RGB data (color value of each pixel) and three kinds of color values (R_w G_w B_w) of each pixel, shown in equation (1).As further instruction, system also can carry out the combination of color conversion and down-sampling, and respectively, the input of system is a rgb format and to export be the YUV420 form.In fact, can make up eight top steps the processing stage in any amount carrying out by system, as long as these combinations the processing stage be the order in succession that is in as shown in Figure 2.As counter-example, system can not carry out the combination between color correction and the down-sampling, because brightness/contrast adjustment and color conversion stage have been walked around in this combination.

One of key component of system of the present invention is determining of the coefficient data that loads in data storage 401.According to top analysis, the complete mapping relations between the input of system and the output can be written as:

$Z(s,t)=f(\underset{m,n}{\overset{M,N}{\mathrm{\Σ}}}{a}_{m,n}^{\′\′}f\left(\underset{k,l}{\overset{K,L}{\mathrm{\Σ}}}{a}_{k,l}^{\′}X(i+k,j+l)\right))---\left(7\right)$

α ' wherein _{K, l}, α " _{M, n}Be respectively a factor of the first order and multiplications/additions unit, the second level, the mapping relations that f () representative is carried out by lookup unit.According at reference book " the application neural net (Applied Neural Networks for signal Processing) that is used for signal processing " (Fa-Long Luo andRolf Unbehauen, Cambrjdge Universe Press, 2000) the relevant approximation theorem that provides in, any Nonlinear Mapping can be approached with arbitrarily small error by the system that can obtain equation (7), and its mapping can by factor alpha ' _{K, l}, α " _{M, n}Determine uniquely.In other words, if specified these coefficients, any mapping and processing between input color and the output color can be finished by system of the present invention.Different different mapping and the processing of coefficient decision.This means by changing these coefficient datas simply, can realize any different coloured image the processing stage and their combination.

Can provide the coefficient data of optimum handling property to pre-determine by use test card vector.As an illustration, how to obtain the coefficient data of the optimum of full tupe.In this case, can optimal coefficient be determined to be written as following optimal problem with the p-norm:

$\min \underset{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="A20061000541900201.png,A20061000541900231.png"\; state="\{\{state\}\}">C</figure-callout>}1}{\mathrm{\&Sigma;}}\&CenterDot;\underset{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20061000541900201.png"\; state="\{\{state\}\}">C</figure-callout>}2}{\mathrm{\&Sigma;}}|\stackrel{\&OverBar;}{Z}(s,t)-f\left(\underset{m,n}{\overset{M,N}{\mathrm{\&Sigma;}}}{a}_{m,n}^{\&prime;\&prime;}f\left(\underset{k,l}{\overset{K,L}{\mathrm{\&Sigma;}}}{a}_{k,l}^{\&prime;}X(i+k,j+l)\right)\right){|}^P---\left(8\right)$

Wherein C1 represents the quantity of all different lighting conditions, and C2 represents the quantity of neighbor related in the color treatments to specified pixel, X (i+k j+l) is the original RGB data of the test card that obtained by transducer, and It is the YUV420 value of the expectation of corresponding test card.

In fact, (i+k, j+l) expression color value is and to different tupes by X

With difference.For example, be that original RGB data (color value of each pixel) and output are the multiprocessing patterns of the RGB data (three color values of each pixel) of interpolation to input, with not being both of full tupe

The rgb format of expression interpolation rather than the YUV420 form of test card.

As mentioned above, to some single tupe, for example from RGB to the color conversion of YUV and the color correction of equation (2), function f () can be a unit gain.This becomes simply the optimal coefficient data of the correspondence of these patterns of acquisition, mainly is because optimization has become linear problem.

Embodiment 2: the main difference of the structure of the graphic system of present embodiment and embodiment 1 is, also comprises multiplexer 801 and demodulation multiplexer 803 as shown in Figure 8; Color of pixel value (X (the i of input, j) to X (i+K, j+L)) behind multiplexer 801, be transferred in the multiplications/additions unit 402, through carry out multiplying with the corresponding coefficient that from data storage 401, is written into and with the result of product addition after, carry out Nonlinear Mapping by lookup unit 403; Repeating under instruction 802 the control by pattern configurations control unit 404, repeating multiplication, addition and Nonlinear Mapping by multiplications/additions unit 402 and lookup unit 403 handles, until the processing of finishing all color of pixel values, then, color of pixel value (Z (s after demodulation multiplexer 803 outputs are handled, t) to Z (s+M, t+N)).

Embodiment 3: about color image processing method of the present invention, the handling process of color of pixel value as shown in Figure 9.At first, determine tupe 902 user's control 901 times according to user instruction.Tupe comprises full tupe, single tupe and multiprocessing pattern.In full tupe, the input of system is original RGB data, and the data of the YUV420 form of system's output down-sampling; In single tupe, system can be configured to carry out following color treatments any in the stage: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling; In the multiprocessing pattern, system can be configured to carry out the more than one color treatments stage.Secondly, the pending pixel value of input is carried out multiplying 903 with the corresponding coefficient value that is written into from data storage according to determined tupe, and with the results added of being taken advantage of.Once more, the result to addition carries out Nonlinear Mapping 904.Then, according to reprocessing instruction 905, determine whether the data after the Nonlinear Mapping are carried out above processing procedure once more.In above-mentioned processing, configuration data memory is written into each and handles required corresponding coefficient.After all processes pixel are intact, the output result.

The color value of wherein treating processed pixels is carried out the processing of multiplication, addition and Nonlinear Mapping, can be one-level, secondary, more than three grades or three grades.One-level described herein is handled and is meant that the color value for the treatment of processed pixels carries out multiplication, addition and a Nonlinear Mapping successively and handles, two stage treatment is the result that the first order is handled to be carried out multiplication, addition and a Nonlinear Mapping more successively handle, more than three grades or three grades then.Wherein said Nonlinear Mapping can be the Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions; It can be by a plurality of multiplier executed in parallel that the described its corresponding coefficient value of color value for the treatment of processed pixels is carried out multiplying, also can be to be repeated by a multiplier; Described Nonlinear Mapping can also can be to be repeated by a lookup unit by a plurality of lookup unit executed in parallel.

More than by specific embodiment the present invention is described in detail, still, it should be noted that at this present invention is not limited to these embodiment, can also do various variations and adjustment to system of the present invention.For example, system configuration of the present invention shown in Figure 4 can also be level Four, Pyatyi etc.Secondly, the multiplications/additions unit in the image processing system of the present invention, lookup unit and pattern configurations control unit can adopt other variety of ways to realize.Data storage can adopt various memory devices, for example various memory devices such as magnetic, light and semiconductor.In addition, also be noted that, employed term notions such as " first, second " is relative in specification of the present invention and claims, only be convenience rather than restriction in order to describe, employed terms such as " more than three grades, a plurality of " is meant that determine and that be of practical significance according to actual needs quantity.In a word, no matter the present invention is made what kind of change and adjustment, only otherwise deviate from spirit of the present invention and claim and equivalent institute restricted portion thereof, all should fall within the protection of the present invention.

Claims (12)

1. color picture processing system is characterized in that comprising:

Data storage is used to store various coefficients required when carrying out different disposal;

The multiplications/additions unit, be used for pending color of pixel value respectively with the corresponding multiplication that is written into from described data storage, and with the results added of being taken advantage of;

Lookup unit is used for Nonlinear Mapping is carried out in the output of described multiplications/additions unit;

The pattern configurations control unit is used for determining the tupe of system's execution, and when carrying out determined tupe coefficient is written into described multiplications/additions unit and described lookup unit from described data storage.

2. color picture processing system according to claim 1 is characterized in that:

Comprise secondary multiplications/additions unit and lookup unit, connect and compose in turn by multiplications/additions unit, lookup unit, multiplications/additions unit and lookup unit; The quantity of each grade multiplication/adder unit and lookup unit is corresponding with the quantity of pending pixel; And each multiplications/additions unit is connected with described pattern configurations control unit with data storage respectively with lookup unit, to receive coefficient data under the control of described pattern configurations control unit.

3. color picture processing system according to claim 1 is characterized in that:

Described multiplications/additions unit and lookup unit can also be one-level, more than three grades or three grades.

4. color picture processing system according to claim 1 is characterized in that:

Also comprise multiplexer and demodulation multiplexer;

Each color of pixel value is transferred to described multiplications/additions unit behind multiplexer, through carry out multiplying with the corresponding coefficient that from described data storage, is written into and with the result of product addition after, carry out Nonlinear Mapping by lookup unit; Under the control that repeats instruction, repeating multiplication, addition and Nonlinear Mapping by described multiplications/additions unit and described lookup unit handles, after the processing of finishing all pixel color value, by the treated color of pixel value of demodulation multiplexer output.

5. according to claim 1,2,3 or 4 described color picture processing systems, it is characterized in that:

Thereby described coefficient is to use a plurality of adjacent pixels predetermined by off-line training.

6. according to claim 1,2,3 or 4 described color picture processing systems, it is characterized in that:

Described multiplications/additions unit is made up of a plurality of multipliers and an accumulator, perhaps is made up of a multiplier and adder and carries out repeatedly multiplication and addition with repetitive mode.

7. according to claim 1,2,3 or 4 described color picture processing systems, it is characterized in that:

Described lookup unit can be a plurality of, or one and repeatedly operation; The content of the table in the described lookup unit and size are predetermined before using;

Described Nonlinear Mapping can be the Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions.

8. according to claim 1,2,3 or 4 described color picture processing systems, it is characterized in that:

Described pattern configurations control unit is determined the tupe that system carries out according to user's instruction, and described tupe comprises full tupe, single tupe and multiprocessing pattern;

In described full tupe, the input of system is original RGB data, and the data of the YUV420 form of system's output down-sampling;

In described single tupe, system can be configured to carry out following color treatments any in the stage: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling;

In described multiprocessing pattern, system can be configured to carry out the more than one color treatments stage.

9. color image processing method is characterized in that may further comprise the steps:

Obtain and store the required various coefficients of execution different disposal in advance by off-line training;

Determine tupe according to user instruction;

Treat its corresponding coefficient of color value of processed pixels according to determined tupe and carry out multiplying, and with the results added of being taken advantage of;

Result to addition carries out Nonlinear Mapping.

10. color image processing method according to claim 9 is characterized in that:

The described color value for the treatment of processed pixels is carried out the processing of multiplication, addition and Nonlinear Mapping, can be one-level, secondary, more than three grades or three grades.

11., it is characterized in that according to claim 9 or 10 described color image processing methods:

Described tupe comprises full tupe, single tupe and multiprocessing pattern;

In described full tupe, the input of system is original RGB data, and the data of the YUV420 form of system's output down-sampling;

In described single tupe, system can be configured to carry out following color treatments any in the stage: defect correction, color interpolation, white balance, Gamma adjustment, color correction, brightness and contrast's enhancing, color conversion and down-sampling;

In described multiprocessing pattern, system can be configured to carry out the more than one color treatments stage;

Described Nonlinear Mapping can be the Gaussian function, piecewise linear function, sigmoid function, thin dish spline function, many quadratic functions, cube approximate or contrary many quadratic functions.

12., it is characterized in that according to claim 9 or 10 described color image processing methods:

It can be by a plurality of multiplier executed in parallel that the described its corresponding coefficient of color value for the treatment of processed pixels is carried out multiplying, also can be to be repeated by a multiplier;

Described Nonlinear Mapping can also can be to be repeated by a lookup unit by a plurality of lookup unit executed in parallel.

Images