CN100563341C - Multiple dimensioned compatible processing devices and methods therefor in image and the video coding - Google Patents

Abstract

The invention discloses multiple dimensioned compatible processing devices and methods therefor in a kind of image and the video coding.It comprises convergent-divergent and the forward processor, the reverse process device that quantize to be separated or to combine.Convergent-divergent comprises the input that is connected successively, multiple dimensioned compatible positive-going transition, quantification and output device with the forward processor that quantizes to be separated, and multiple dimensioned compatible device for zooming is arranged between multiple dimensioned compatible positive-going transition and the quantization device.Convergent-divergent comprises the input that is connected successively, multiple dimensioned compatible positive-going transition and output device with the forward processor that quantizes to combine, and multiple dimensioned compatible scalar quantization device is arranged between multiple dimensioned compatible positive-going transition and the output device.The reverse process device comprises the input that is connected successively, multiple dimensioned compatible transformation by reciprocal direction and output device.It has characteristics simple in structure, stable performance, and it can carry out multiple dimensioned compatible processing to video, image or residual signals.Need not additional calculations or hardware spending and can realize multi-scale transform, convergent-divergent and quantification simultaneously.

Description

Multiple dimensioned compatible processing devices and methods therefor in image and the video coding

Affiliated technical field

The invention belongs to the image communication technology field, particularly relate to multiple dimensioned compatible processing devices and methods therefor in a kind of image and the video coding.

Background technology

In the epoch that image communication technology develops rapidly, data compression technique is to realize data preservation and one of data transfer key in high quality, people are to the research of video compression technology and use and also had more than 20 year, produced in the communications field H.261, H.263, H.264, MEPG-1, MPEG-2, MPEG-4 and domestic standard A VS video encoding standard that first is formulated at the audio frequency and video industry demand.In actual applications, often some kinds of methods are mixed and use, take all factors into consideration prediction, the coding framework of conversion and entropy coding in the hope of obtaining higher compression ratio, has following main feature:

1, utilizes inter prediction to come redundancy between removal of images, utilize infra-frame prediction to come redundancy in the removal of images.

2, come visual redundancy in the removal of images by prediction residual being carried out transform and quantization.

3, utilize entropy coding to eliminate statistical redundancy.

Above-mentioned video encoding standard all has in-frame encoding picture frame (I frame) and inter coded images frame (P frame), and the I frame adopts different coding methods with the P frame.The cataloged procedure of I frame is as follows: can adopt the infra-frame prediction based on frequency domain or spatial domain, reference value preparation method with the current block to be encoded of predictive mode mark, calculate the difference between block to be encoded and its reference block then, obtain corresponding residual error data piece, raw image data piece or residual error data piece are carried out 2 dimension conversion, in transform domain, conversion coefficient is quantized then, carry out entropy coding at last, promptly adopt variable-length encoding or arithmetic coding etc.The cataloged procedure of P frame is as follows: adopt the inter prediction based on motion compensation, come the reference block locations of the current block to be encoded of mark by motion vector, calculate the difference between block to be encoded and its reference block then, obtain corresponding residual error data piece, then the residual error data piece carries out 2 dimension conversion, again coefficient in transform domain is quantized, carry out entropy coding at last.

In video coding, coding side only needs the change quantization coefficient of transfer encoding pattern information, motion vector, reconstructed image piece and its predicted value residual error to get final product.And, received coding mode information and motion vector in decoding end, and can current decoding block be predicted that obtain its predicted value, the change quantization coefficient to residual error carries out the inverse quantization inverse transformation again, can obtain the reconstructed image piece with the predicted value addition.

Traditional image encoding standard as Joint Photographic Experts Group, has close part with video encoding standard, and the residual block that prediction in raw image data or the image is obtained carries out two-dimensional transform; In transform domain, conversion coefficient is quantized then; Carry out entropy coding at last.

Because video data and view data stronger correlation on spatial domain, two-dimensional transform is the key factor that improves coding gain, so two-dimensional transform is the very part and parcel of video coding and image encoding.

Discrete cosine transform (DCT) conversion or the integer transform that approaches discrete cosine transform are generally used for the block transform coding of view data and video data, this is because for various signals, and discrete cosine transform and integer transform are similar to the best Karhunen-Loeve transformation of statistics very much.Discrete cosine transform and integer transform are widely used in the various video/image coding standards.In existing video/image coding standard, part of standards adopts 4 * 4 conversion, convergent-divergent and quantization method that colourity/luminance block is all carried out 4 * 4 conversion, convergent-divergent and quantification treatment, and part of standards adopts 8 * 8 conversion, convergent-divergent and quantization method that colourity/luminance block is all carried out 8 * 8 conversion, convergent-divergent and quantification treatment.Adopt 4 * 4 conversion, convergent-divergent and quantization method amount of calculation little, but it is also little to remove the signal correction sexuality, and adopt 8 * 8 conversion, convergent-divergent and quantization method big than 4 * 4 conversion, convergent-divergent and quantization method amount of calculation, it is big to remove the signal correction sexuality, but carry out 8 * 8 conversion, convergent-divergent and quantification for 8 * 8 blocks of data that are spliced by 44 * 4, its poor effect.

Original picture block or residual block for 2m * 2n, 2m * n, m * 2n, m * n, should adopt conversion, convergent-divergent and the quantification of 2m * 2n, 2m * n, m * 2n, m * n to handle respectively, but in traditional treatment system, this need introduce 4 cover processing unit and methods, complexity is very high, has influenced the feasibility that realizes.Wherein, m and n can be respectively big or small arbitrarily positive integers.

Summary of the invention

The object of the present invention is to provide a kind of simple in structure, stable performance, the predicted residual signal of video, image, video or image is carried out multiple dimensioned compatible processing devices and methods therefor.

In order to achieve the above object, the technical solution used in the present invention is as follows:

A kind of technical scheme of multiple dimensioned compatible processing device is achieved in that it comprises convergent-divergent and the forward processor that quantizes to be separated in image of the present invention and the video coding, described convergent-divergent comprises the input unit that is connected successively, multiple dimensioned compatible positive-going transition device, quantization device and output device with the forward processor that quantizes to be separated, and it is characterized in that: described multiple dimensioned compatible device for zooming is arranged between multiple dimensioned compatible positive-going transition device and the quantization device.

A kind of method of multiple dimensioned compatible processing is convergent-divergent and the forward processing method that quantizes to be separated in image of the present invention and the video coding, it is characterized in that the processing method of multiple dimensioned compatible device for zooming:

Multiple dimensioned compatible positive-going transition device is exported the f that obtains _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible convergent-divergent and handle, finish just reverse convergent-divergent and handle, as follows:

${\mathrm{sf}}_{2m\×2n}=f_{2m\×2n}\⊗S_{2m\×2n},$ $s{f}_{2m\×n}=f_{2m\×n}\⊗S_{2m\×n},$ ${\mathrm{sf}}_{m\×2n}=f_{m\×2n}\⊗S_{m\×2n}$ Or ${\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗S_{m\×n},$

Wherein, sf _{2m * 2n}, sf _{2m * n}, sf _{M * 2n}Or sf _{M * n}Represent the multiple dimensioned compatible convergent-divergent result of 2m * 2n, 2m * n, m * 2n or m * n respectively, operator

Expression is multiplied each other S respectively with the coefficient of same position in two matrixes about it _{2m * 2n}, S _{2m * n}, S _{M * 2n}Or S _{M * n}Represent the multiple dimensioned compatible zoom factor matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, S _{2m * 2n}In (i, j) individual coefficient is S _{2m * 2n}(i, j)=1/ (s _i* s _j) ², s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, S _{2m * n}(i, j)=2 * S _{2m * 2n}(i, 2j), S _{M * 2n}(i, j)=2 * S _{2m * 2n}(2i, j), S _{M * n}(i, j)=4 * S _{2m * 2n}(2i, 2j).

Utilize 2m * 2n integer zoom factor matrix S I _{2m * 2n}Structure 2m * n, m * 2n or m * n integer zoom factor matrix S I _{2m * n}, SI _{M * 2n}Or SI _{M * n}, as follows:

SI _{2m * n}(i, j)=SI _{2m * 2n}(i, 2j), SI _{M * 2n}(i, j)=SI _{2m * 2n}(2i, j) or SI _{M * n}(i, j)=SI _{2m * 2n}(2i, 2j),

When carrying out right-shift operation, 2m * n convergent-divergent or m * 2n convergent-divergent moves to right 1 less with respect to 2m * 2n convergent-divergent, and m * n convergent-divergent moves to right 2 less with respect to 2m * 2n convergent-divergent.

The another kind of technical scheme of multiple dimensioned compatible processing device is achieved in that it comprises convergent-divergent and the forward processor that quantizes to combine in image of the present invention and the video coding, described convergent-divergent comprises the input unit that is connected successively, multiple dimensioned compatible positive-going transition device and output device with the forward processor that quantizes to combine, and it is characterized in that: described multiple dimensioned compatible scalar quantization device is arranged between multiple dimensioned compatible positive-going transition device and the output device.

The another kind of method of multiple dimensioned compatible processing is convergent-divergent and the forward processing method that quantizes to combine in image of the present invention and the video coding, it is characterized in that the processing method of multiple dimensioned compatible scalar quantization device:

Multiple dimensioned compatible positive-going transition device is exported the f that obtains _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible scalar quantization and handle, finish just reverse scalar quantization and handle, as follows:

${\mathrm{qsf}}_{2m\×2n}=f_{2m\×2n}\⊗{\mathrm{QS}}_{2m\×2n},$ ${\mathrm{qsf}}_{2m\×n}=f_{2m\×n}\⊗{\mathrm{QS}}_{2m\×n},$ ${\mathrm{qsf}}_{m\×2n}=f_{m\×2n}\⊗{\mathrm{QS}}_{m\×2n}$ Or

${\mathrm{qsf}}_{m\×n}=f_{m\×n}\⊗Q{S}_{m\×n},$

Wherein, qsf _{2m * 2n}, qsf _{2m * n}, qsf _{M * 2n}Or qsf _{M * n}Represent the multiple dimensioned compatible scalar quantization result of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}, QS _{2m * n}, QS _{M * 2n}Or QS _{M * n}Represent the multiple dimensioned compatible scalar quantization coefficient matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}In (i, j) individual coefficient is QS _{2m * 2n}(i, and j)=1/ (Q (i, j) * (s _i* s _j) ²), s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, (i j) is the quantization step factor, QS to Q _{2m * n}(i, j)=2 * QS _{2m * 2n}(i, 2j), QS _{M * 2n}(i, j)=2 * QS _{2m * 2n}(2i, j), QS _{M * n}(i, j)=4 * QS _{2m * 2n}(2i, 2j).

Utilize 2m * 2n integer scalar quantization coefficient matrix QSI _{2m * 2n}Structure 2m * n, m * 2n or m * n integer scalar quantization coefficient matrix QSI _{2m * n}, QSI _{M * 2n}Or QSI _{M * n}, as follows:

QSI _{2m * n}(i, j)=QSI _{2m * 2n}(i, 2j), QSI _{M * 2n}(i, j)=QSI _{2m * 2n}(2i, j) or QSI _{M * n}(i, j)=QSI _{2m * 2n}(2i, 2j),

When carrying out the right-shift operation under the same degree quantification, 2m * n scalar quantization or m * 2n scalar quantization moves to right 1 less with respect to 2m * 2n scalar quantization, and m * n scalar quantization moves to right 2 less with respect to 2m * 2n scalar quantization.

The another kind of technical scheme of multiple dimensioned compatible processing device is achieved in that it comprises the reverse process device in image of the present invention and the video coding, it is characterized in that described reverse process device is made of input unit, multiple dimensioned compatible transformation by reciprocal direction device and output device, is connected with multiple dimensioned compatible transformation by reciprocal direction device between input unit and output device.

The another kind of method of multiple dimensioned compatible processing is the transformation by reciprocal direction method in image of the present invention and the video coding, it is characterized in that this transformation by reciprocal direction method step is:

1) chooses transformation by reciprocal direction matrix size and coefficient according to positive-going transition matrix size and coefficient;

2) according to set transformation by reciprocal direction matrix the input data being carried out transformation by reciprocal direction handles.

The present invention compares with background technology, the beneficial effect that has:

On the basis that the present invention combines forward convergent-divergent and reverse convergent-divergent when positive-going transition is handled, utilize 2m * 2n, 2m * n, m * 2n or m * n conversion, the consistency of convergent-divergent and quantization operations, the operation and the arithmetic unit that can synthesize a kind of compatibility, need not any additional calculations or hardware spending and just can realize 2m * 2n simultaneously, 2m * n, m * 2n or m * n conversion, convergent-divergent and quantification, according to the view data of input or the block size and the characteristics of video data, reasonably select alternative types, thereby have the computing of minimizing storage complexity, reduce cost, improve advantages such as signal pressure shrinkage.The present invention is not limited to image encoding and field of video encoding, can extend in the application of other signal processing.The present invention simultaneously has certain extensibility, can combine with numerous signal processing methods, and can realize the signal processing of different application scope.

Description of drawings

Fig. 1 is the convergent-divergent and the forward processor structure block diagram that quantizes to be separated of multiple dimensioned compatible processing device of the present invention;

Fig. 2 is the convergent-divergent and the forward processor structure block diagram that quantizes to combine of multiple dimensioned compatible processing device of the present invention;

Fig. 3 is the reverse process device structured flowchart of multiple dimensioned compatible processing device of the present invention.

Embodiment

The pieces that view data and video data or its residual error data can be divided into different sizes carry out conversion.For example, 2m * 2n, 2m * n, m * 2n or m * n etc.Piece for these different sizes can carry out 2m * 2n, 2m * n, m * 2n or m * n conversion, convergent-divergent and quantification respectively.

With reference to accompanying drawing 1: the convergent-divergent of multiple dimensioned compatible processing device of the present invention comprises the input unit 110 that is connected successively, multiple dimensioned compatible positive-going transition device 120, multiple dimensioned compatible device for zooming 130, quantization device 140 and output device 150 with the forward processor that quantizes to be separated; Described multiple dimensioned compatible device for zooming 130 is arranged between multiple dimensioned compatible positive-going transition device 120 and the quantization device 140.

Input unit 110 adopt existing volatile memories (as register, high-speed cache, RAM), nonvolatile storage (as ROM, EEPROM, flash memory) or both certain combine, it is used to accept to store the input data of multiple dimensioned compatible positive-going transition, these data can be vision signals, picture signal, the predicted residual signal of video or image etc., it can be accepted to import data and keep, and when needs carry out multiple dimensioned compatible positive-going transition and handle, can provide the input data.

Multiple dimensioned compatible positive-going transition device 120 couples of 2m * 2n, 2m * n, m * 2n or m * n input block carry out multiple dimensioned compatible positive-going transition to be handled, comprising ALU 125 can realize this processing with multiplying, add operation and shift operation.Input signal has become transform-domain signals by the spatial domain signal transition behind multiple dimensioned compatible positive-going transition device 120, the big correlation that has between the signal of spatial domain greatly reduces in frequency-region signal.

Multiple dimensioned compatible device for zooming 130 couples of 2m * 2n, 2m * n, m * 2n or m * n transform-domain signals carry out multiple dimensioned compatible convergent-divergent to be handled, comprising ALU 135 can realize this processing with division arithmetic, multiplying, add operation and shift operation.Transform-domain signals has been realized after the positive-going transition and the processing of the convergent-divergent before the transformation by reciprocal direction by behind the multiple dimensioned compatible device for zooming 130 simultaneously, and signal has been realized after the positive-going transition and the normalized before the transformation by reciprocal direction.

The dateout of 140 pairs of multiple dimensioned compatible device for zooming 130 of quantization device is carried out quantification treatment, realizes diminishing or lossless compress.

Output device 150 is used to store the data after the quantification treatment, and output is provided, and is similar to input unit 110, when needs are exported, can provide result data.

With reference to accompanying drawing 2: the convergent-divergent of multiple dimensioned compatible processing device of the present invention comprises the input unit 210 that is connected successively, multiple dimensioned compatible positive-going transition device 220, multiple dimensioned compatible scalar quantization device 230 and output device 240 with the forward processor that quantizes to combine; Described multiple dimensioned compatible scalar quantization device 230 is arranged between multiple dimensioned compatible positive-going transition device 220 and the output device 240.

Input unit 210 adopt existing volatile memories (as register, high-speed cache, RAM), nonvolatile storage (as ROM, EEPROM, flash memory) or both certain combine, it is used to accept to store the input data of multiple dimensioned compatible positive-going transition, these data can be vision signals, picture signal, the predicted residual signal of video or image etc., it can be accepted to import data and keep, and when needs carry out multiple dimensioned compatible positive-going transition and handle, can provide the input data.

Multiple dimensioned compatible positive-going transition device 220 couples of 2m * 2n, 2m * n, m * 2n or m * n input block carry out multiple dimensioned compatible positive-going transition to be handled, comprising ALU 225 can realize this processing with multiplying, add operation and shift operation.Input signal has become transform-domain signals by the spatial domain signal transition behind multiple dimensioned compatible positive-going transition device 220, the big correlation that has between the signal of spatial domain greatly reduces in frequency-region signal.

Multiple dimensioned compatible scalar quantization device 230 couples of 2m * 2n, 2m * n, m * 2n or m * n transform-domain signals carry out multiple dimensioned compatible scalar quantization to be handled, comprising ALU 235 can realize this processing with division arithmetic, multiplying, add operation and shift operation.Transform-domain signals has been realized after the positive-going transition and the convergent-divergent before the transformation by reciprocal direction is handled and quantification treatment after by multiple dimensioned compatible scalar quantization device 230 simultaneously, and signal has been realized after the positive-going transition and normalized and lossy compression method before the transformation by reciprocal direction.

Output device 240 is used to store the data after multiple dimensioned compatible scalar quantization is handled, and output is provided, and is similar to input unit 210, when needs are exported, can provide result data.

With reference to accompanying drawing 3: the reverse process device of the multiple dimensioned compatible processing device of the present invention is made of input unit 310, multiple dimensioned compatible transformation by reciprocal direction device 320 and output device 330, is connected with multiple dimensioned compatible transformation by reciprocal direction device 320 between input unit 310 and output device 330.

2m * 2n, 2m * n, m * 2n or m * n that described input unit 310 is used to accept to store multiple dimensioned compatible transformation by reciprocal direction imports data; Multiple dimensioned compatible transformation by reciprocal direction device 320 couples of 2m * 2n, 2m * n, m * 2n or m * n input block carry out multiple dimensioned compatible transformation by reciprocal direction to be handled, comprising ALU 325 can realize this processing with multiplying, add operation and shift operation; Output device 330 is used to store the data behind the transformation by reciprocal direction, and output is provided.

Input unit 310, multiple dimensioned compatible transformation by reciprocal direction device 320 and output device 330 all are similar to corresponding part in convergent-divergent and the forward processor that forward processor that quantizes to be separated or convergent-divergent combine with quantification, input unit 310 can be stored the normalization transform-domain signals, behind multiple dimensioned compatible transformation by reciprocal direction device 320, change original or approximate original video, image or video image residual signals into.

Above-mentioned convergent-divergent can be used processor system with the forward processor that quantizes to be separated, forward processor and the multiple dimensioned compatible flyback converter that convergent-divergent combines with quantification, microcontroller, programmable logic device or the part or all of operation of microprocessor realization.Above-mentioned certain operations can realize that other operations simultaneously can realize with hardware, perhaps all realize with hardware with software.

For convenience's sake, these operations are described to the functional unit of different interconnection or different module of software and hardware.But this is dispensable.In some applications, these functional units or module can be integrated into single logical device, in program or the operation, and do not have obvious limit.In any situation, the feature of functional unit and module of software and hardware or description can independently realize, or reinstate hardware or software realization with other operations one.

A kind of method of multiple dimensioned compatible processing is convergent-divergent and the forward processing method that quantizes to be separated, and the steps include:

1) multiple dimensioned compatible positive-going transition is carried out in one or several combination in 2m * 2n, 2m * n, m * 2n or m * n input signal respectively and handled, as follows:

f _{2m * 2n}=T _2m* F _{2m * 2n}* T _2n, f _{2m * n}=T _2m* F _{2m * n}* T _n, f _{M * 2n}=T _m* F _{M * 2n}* T _2nOr f _{M * n}=T _m* F _{M * n}* T _n,

Wherein, F _{2m * 2n}, F _{2m * n}, F _{M * 2n}Or F _{M * n}Represent the multiple dimensioned compatible positive-going transition input data of 2m * 2n, 2m * n, m * 2n or m * n respectively, f _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}Represent the multiple dimensioned compatible positive-going transition result of 2m * 2n, 2m * n, m * 2n or m * n respectively, T _2m, T _m, T _2nAnd T _nBe respectively 2m * 2m, m * m, 2n * 2n and n * n transformation matrix, and T _2mAnd T _m, T _2nAnd T _nSatisfy following relation:

$T_{2m}=\left[\begin{array}{cc}{T}_m& 0\\ 0& T_m^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_m& I_m\\ I_m& -I_m\end{array}\right],$ $T_{2n}=\left[\begin{array}{cc}{T}_n& 0\\ 0& T_n^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_n& I_n\\ I_n& -I_n\end{array}\right],$

T ' wherein _mAnd T ' _nBe respectively m * m and n * n matrix, I _mAnd I _nIt is respectively m * m and n * n unit matrix;

2) f to obtaining after the multiple dimensioned compatible positive-going transition processing _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible convergent-divergent and handle, finish just reverse convergent-divergent and handle, as follows:

${\mathrm{sf}}_{2m\×2n}=f_{2m\×2n}\⊗S_{2m\×2n},$ ${\mathrm{sf}}_{2m\×n}=f_{2m\×n}\⊗S_{2m\×n},$ ${\mathrm{sf}}_{m\×2n}=f_{m\×2n}\⊗S_{m\×2n}$ Or ${\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗S_{m\×n},$

Wherein, sf _{2m * 2n}, sf _{2m * n}, sf _{M * 2n}Or sf _{M * n}Represent the multiple dimensioned compatible convergent-divergent result of 2m * 2n, 2m * n, m * 2n or m * n respectively, operator

Expression is multiplied each other S respectively with the coefficient of same position in two matrixes about it _{2m * 2n}, S _{2m * n}, S _{M * 2n}Or S _{M * n}Represent the multiple dimensioned compatible zoom factor matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, S _{2m * 2n}In (i, j) individual coefficient is S _{2m * 2n}(i, j)=1/ (s _i* s _j) ², s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, S _{2m * n}(i, j)=2 * S _{2m * 2n}(i, 2j), S _{M * 2n}(i, j)=2 * S _{2m * 2n}(2i, j), S _{M * n}(i, j)=4 * S _{2m * 2n}(2i, 2j);

3) sf to obtaining after the multiple dimensioned compatible convergent-divergent processing _{2m * 2n}, sf _{2m * n}, sf _{M * 2n}Or sf _{M * n}In one or several combination carry out quantification treatment.

Utilize 2m * 2n integer zoom factor matrix S I _{2m * 2n}Structure 2m * n, m * 2n or m * n integer zoom factor matrix S I _{2m * n}, SI _{M * 2n}Or SI _{M * n}, as follows:

SI _{2m * n}(i, j)=SI _{2m * 2n}(i, 2j), SI _{M * 2n}(i, j)=SI _{2m * 2n}(2i, j) or SI _{M * n}(i, j)=SI _{2m * 2n}(2i, 2j),

When carrying out right-shift operation, 2m * n convergent-divergent or m * 2n convergent-divergent moves to right 1 less with respect to 2m * 2n convergent-divergent, and m * n convergent-divergent moves to right 2 less with respect to 2m * 2n convergent-divergent.

The another kind of method of multiple dimensioned compatible processing is convergent-divergent and the forward processing method that quantizes to combine, and the steps include:

1) multiple dimensioned compatible positive-going transition is carried out in one or several combination in 2m * 2n, 2m * n, m * 2n or m * n input signal respectively and handled, as follows:

f _{2m * 2n}=T _2m* F _{2m * 2n}* T _2n, f _{2m * n}=T _2m* F _{2m * n}* T _n, f _{M * 2n}=T _m* F _{M * 2n}* T _2nOr f _{M * n}=T _m* F _{M * n}* T _n,

Wherein, F _{2m * 2n}, F _{2m * n}, F _{M * 2n}Or F _{M * n}Represent the multiple dimensioned compatible positive-going transition input data of 2m * 2n, 2m * n, m * 2n or m * n respectively, f _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}Represent the multiple dimensioned compatible positive-going transition result of 2m * 2n, 2m * n, m * 2n or m * n respectively, T _2m, T _m, T _2nAnd T _nBe respectively 2m * 2m, m * m, 2n * 2n and n * n transformation matrix, and T _2mAnd T _m, T _2nAnd T _nSatisfy following relation:

$T_{2m}=\left[\begin{array}{cc}{T}_m& 0\\ 0& T_m^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_m& I_m\\ I_m& {-I}_m\end{array}\right],$ $T_{2n}=\left[\begin{array}{cc}{T}_n& 0\\ 0& T_n^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_n& I_n\\ I_n& {-I}_n\end{array}\right],$

T ' wherein _mAnd T ' _nBe respectively m * m and n * n matrix, I _mAnd I _nIt is respectively m * m and n * n unit matrix;

2) f to obtaining after the multiple dimensioned compatible positive-going transition processing _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible scalar quantization and handle, finish just reverse scalar quantization and handle, as follows:

${\mathrm{qsf}}_{2m\×2n}=f_{2m\×2n}\⊗{\mathrm{QS}}_{2m\×2n},$ ${\mathrm{qsf}}_{2m\×n}=f_{2m\×n}\⊗{\mathrm{QS}}_{2m\×n},$ ${\mathrm{qsf}}_{m\×2n}=f_{m\×2n}\⊗{\mathrm{QS}}_{m\×2n}$ Or

${\mathrm{qsf}}_{m\×n}=f_{m\×n}\⊗{\mathrm{QS}}_{m\×n},$

Wherein, qsf _{2m * 2n}, qsf _{2m * n}, qsf _{M * 2n}Or qsf _{M * n}Represent the multiple dimensioned compatible scalar quantization result of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}, QS _{2m * n}, QS _{M * 2n}Or QS _{M * n}Represent the multiple dimensioned compatible scalar quantization coefficient matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}In (i, j) individual coefficient is QS _{2m * 2n}(i, and j)=1/ (Q (i, j) * (s _i* s _j) ²), s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, (i j) is the quantization step factor, QS to Q _{2m * n}(i, j)=2 * QS _{2m * 2n}(i, 2j), QS _{M * 2n}(i, j)=2 * QS _{2m * 2n}(2i, j), QS _{M * n}(i, j)=4 * QS _{2m * 2n}(2i, 2j).

Utilize 2m * 2n integer scalar quantization coefficient matrix QSI _{2m * 2n}Structure 2m * n, m * 2n or m * n integer scalar quantization coefficient matrix QSI _{2m * n}, QSI _{M * 2n}Or QSI _{M * n}, as follows:

QSI _{2m * n}(i, j)=QSI _{2m * 2n}(i, 2j), QSI _{M * 2n}(i, j)=QSI _{2m * 2n}(2i, j) or QSI _{M * n}(i, j)=QSI _{2m * 2n}(2i, 2j),

When carrying out the right-shift operation under the same degree quantification, 2m * n scalar quantization or m * 2n scalar quantization moves to right 1 less with respect to 2m * 2n scalar quantization, and m * n scalar quantization moves to right 2 less with respect to 2m * 2n scalar quantization.

The another kind of method of multiple dimensioned compatible processing is the transformation by reciprocal direction method in image and the video coding, the steps include:

1) chooses transformation by reciprocal direction matrix size and coefficient according to positive-going transition matrix size and coefficient;

2) according to set transformation by reciprocal direction matrix the input data being carried out transformation by reciprocal direction handles.

The example of multiple dimensioned compatibility processing method:

8 * 8 transformation matrixs are:

${\mathrm{ICT}}_8=\left[\begin{array}{cccccccc}8& 8& 8& 8& 8& 8& 8& 8\\ 10& 9& 6& 2& -2& -6& -9& -10\\ 10& 4& -4& -10& -10& -4& 4& 10\\ 9& -2& -10& -6& 6& 10& 2& -9\\ 8& -8& -8& 8& 8& -8& -8& 8\\ 6& -10& 2& 9& -9& -2& 10& -6\\ 4& -10& 10& -4& -4& 10& -10& 4\\ 2& -6& 9& -10& 10& -9& 6& -2\end{array}\right].$

With 4 * 4 transformation matrixs of this matrix compatibility be:

${\mathrm{ICT}}_4=\left[\begin{array}{cccc}8& 8& 8& 8\\ 10& 4& -4& -10\\ 8& -8& -8& 8\\ 4& -10& 10& -4\end{array}\right].$

Multiple dimensioned compatible positive-going transition is carried out in one or several combination in 8 * 8,8 * 4,4 * 8 or 4 * 4 input signals of 9 bits respectively handled, as follows:

f _{8 * 8}=(T ₈* F _{8 * 8}* T ₈+ 16)＞＞5, f _{8 * 4}=(T ₈* F _{8 * 4}* T ₄+ 16)＞＞5, f _{4 * 8}=(T ₄* F _{4 * 8}* T ₈+ 16)＞＞5 or

f _4×4＝(T ₄×F _4×4×T ₄+16)＞＞5，

Wherein,＜＜represent coefficients all in the matrix is all moved to left,＞＞represent coefficients all in the matrix is all moved to right.The convergent-divergent of data is handled and be multiply by zoom factor matrix S caleMatrix according to its position after 8 * 8 conversion _{8 * 8}In the zoom factor of relevant position 19 bits that move to right again, the convergent-divergent of data is handled and be multiply by zoom factor matrix S caleMatrix according to its position after 8 * 4 conversion _{8 * 4}In the zoom factor of relevant position 18 bits that move to right again, the convergent-divergent of data is handled and be multiply by zoom factor matrix S caleMatrix according to its position after 4 * 8 conversion _{4 * 8}In the zoom factor of relevant position 18 bits that move to right again, the convergent-divergent of data is handled and be multiply by zoom factor matrix S caleMatrix according to its position after 4 * 4 conversion _{4 * 4}In the zoom factor of relevant position 17 bits that move to right again, ScaleMatrix _{8 * 8}As follows:

${\mathrm{ScaleMatrix}}_{8\×8}=\left[\begin{array}{cccccccc}32768& 37958& 36158& 37958& 32768& 37958& 36158& 37958\\ 37958& 43969& 41884& 43969& 37958& 43969& 41884& 43969\\ 36158& 41884& 39898& 41884& 36158& 41884& 39898& 41884\\ 37958& 43969& 41884& 43969& 37958& 43969& 41884& 43969\\ 32768& 37958& 36158& 37958& 32768& 37958& 36158& 37958\\ 37958& 43969& 41884& 43969& 37958& 43969& 41884& 43969\\ 36158& 41884& 39898& 41884& 36158& 41884& 39898& 41884\\ 37958& 43969& 41884& 43969& 37958& 43969& 41884& 43969\end{array}\right].$

ScaleMatrix _{8 * 4}In (i, j) individual coefficient S caleMatrix _{8 * 4}(i, j) as follows:

ScaleMatrix _8×4(i，j)＝ScaleMatrix _8×8(i，2j)。

ScaleMatrix _{4 * 8}In (i, j) individual coefficient S caleMatrix _{4 * 8}(i, j) as follows:

ScaleMatrix _4×8(i，j)＝ScaleMatrix _8×8(2i，j)。

ScaleMatrix _{4 * 4}In (i, j) individual coefficient S caleMatrix _{4 * 4}(i, j) as follows:

ScaleMatrix _4×4(i，j)＝ScaleMatrix _8×8(2i，2j)。

Can adopt one group or several groups of computings in cover arithmetic unit realization 8 * 8,8 * 4,4 * 8 or 4 * 4 conversion and the convergent-divergent.Obtain after multiple dimensioned compatible convergent-divergent handled finish after the positive-going transition and the data of the normalized before the transformation by reciprocal direction are carried out quantification treatment, realize diminishing or lossless compress.

In decoding end, choose 8 * 8,8 * 4,4 * 8 or 4 * 4 transformation by reciprocal direction matrix coefficients according to 8 * 8,8 * 4,4 * 8 or 4 * 4 positive-going transition matrix coefficients, according to 8 * 8,8 * 4,4 * 8 or 4 * 4 set transformation by reciprocal direction matrixes 8 * 8,8 * 4,4 * 8 or 4 * 4 input data are carried out multiple dimensioned compatible transformation by reciprocal direction then and handle, recover to obtain original or approximate original video, image or video image residual signals.

Same, also can adopt one group or several groups of computings in cover arithmetic unit realization 8 * 8,8 * 4,4 * 8 or 4 * 4 conversion and the scalar quantization.

Claims (6)

1. multiple dimensioned compatible processing device in image and the video coding, comprise convergent-divergent and the forward processor that quantizes to be separated, described convergent-divergent comprises input unit (110), multiple dimensioned compatible positive-going transition device (120), quantization device (140) and the output device (150) that is connected successively with the forward processor that quantizes to be separated, and it is characterized in that: described multiple dimensioned compatible device for zooming (130) is arranged between multiple dimensioned compatible positive-going transition device (120) and the quantization device (140);

Described multiple dimensioned compatible positive-going transition device (120) carries out multiple dimensioned compatible positive-going transition to 2m * 2n, 2m * n, m * 2n or m * n input block to be handled: f _{2m * 2n}=T _2m* F _{2m * 2n}* T _2n, f _{2m * n}=T _2m* F _{2m * n}* T _n, f _{M * 2n}=T _m* F _{M * 2n}* T _2nOr f _{M * n}=T _m* F _{M * n}* T _n, wherein, F _{2m * 2n}, F _{2m * n}, F _{M * 2n}Or F _{M * n}Represent the multiple dimensioned compatible positive-going transition input data of 2m * 2n, 2m * n, m * 2n or m * n respectively, f _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}Represent the multiple dimensioned compatible positive-going transition result of 2m * 2n, 2m * n, m * 2n or m * n respectively, T _2m, T _m, T _2nWith Tn be respectively 2m * 2m, m * m, 2n * 2n and n * n transformation matrix, and T _2mAnd T _m, T _2nAnd T _nSatisfy following relation:

$T_{2m}=\left[\begin{array}{cc}{T}_m& 0\\ 0& T_m^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_m& I_m\\ I_m& -I_m\end{array}\right],$ $T_{2n}=\left[\begin{array}{cc}{T}_n& 0\\ 0& T_n^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_n& I_n\\ I_n& -I_n\end{array}\right],$

T ' wherein _mAnd T ' _nBe respectively m * m and n * n matrix, I _mAnd I _nIt is respectively m * m and n * n unit matrix; ALU (125) included in the described multiple dimensioned compatible positive-going transition device (120) is realized multiple dimensioned compatible positive-going transition processing with multiplying, add operation and shift operation;

Described multiple dimensioned compatible device for zooming (130) carries out multiple dimensioned compatible convergent-divergent to 2m * 2n, 2m * n, m * 2n or m * n transform-domain signals to be handled: ${\mathrm{sf}}_{2m\×2n}=f_{2m\×2n}\⊗S_{2m\×2n},{\mathrm{sf}}_{2m\×n}=f_{2m\×n}\⊗S_{2m\×n},$ ${\mathrm{sf}}_{m\×2n}=f_{m\×2n}\⊗S_{m\×2n}$ Or ${\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗S_{m\×n},$ Wherein, sf _{2m * 2n}, sf _{2m * n}, sf _{M * 2n}Or sf _{M * n}Represent the multiple dimensioned compatible convergent-divergent result of 2m * 2n, 2m * n, m * 2n or m * n respectively, operator

Expression is multiplied each other S respectively with the coefficient of same position in two matrixes about it _{2m * 2n}, S _{2m * n}, S _{M * 2n}Or S _{M * n}Represent the multiple dimensioned compatible zoom factor matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, S _{2m * 2n}In (i, j) individual coefficient is S _{2m * 2n}(i, j)=1/ (s _i* s _j) ², s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, S _{2m * n}(i, j)=2 * S _{2m * 2n}(i, 2j), S _{M * 2n}(i, j)=2 * S _{2m * 2n}(2i, j), S _{M * n}(i, j)=4 * S _{2m * 2n}(2i, 2j); ALU (135) included in the described multiple dimensioned compatible device for zooming (130) is realized multiple dimensioned compatible convergent-divergent processing with division arithmetic, multiplying, add operation and shift operation.

2. the processing method of multiple dimensioned compatible processing device in a kind of image according to claim 1 and the video coding is characterized in that:

Multiple dimensioned compatible positive-going transition device is exported the f that obtains _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible convergent-divergent and handle, finish just reverse convergent-divergent and handle, as follows:

${\mathrm{sf}}_{2m\×2n}=f_{2m\×2n}\⊗S_{2m\×2n},{\mathrm{sf}}_{2m\×n}=f_{2m\×n}\⊗S_{2m\×n},{\mathrm{sf}}_{m\×2n}=f_{m\×2n}\⊗S_{m\×2n}$ Or ${\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗S_{m\×n},$

Wherein, sf _{2m * 2n}, sf _{2m * n}, sf _{M * 2n}Or sf _{M * n}Represent the multiple dimensioned compatible convergent-divergent result of 2m * 2n, 2m * n, m * 2n or m * n respectively, operator

Expression is multiplied each other S respectively with the coefficient of same position in two matrixes about it _{2m * 2n}, S _{2m * n}, S _{M * 2n}Or S _{M * n}Represent the multiple dimensioned compatible zoom factor matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, S _{2m * 2n}In (i, j) individual coefficient is S _{2m * 2n}(i, j)=1/ (s _i* s _j) ², s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, S _{2m * n}(i, j)=2 * S _{2m * 2n}(i, 2j), S _{M * 2n}(i, j)=2 * S _{2m * 2n}(2i, j), S _{M * n}(i, j)=4 * S _{2m * 2n}(2i, 2j).

3. multiple dimensioned compatibility processing method in a kind of image according to claim 2 and the video coding is characterized in that: utilize 2m * 2n integer zoom factor matrix S I _{2m * 2n}Structure 2m * n, m * 2n or m * n integer zoom factor matrix S I _{2m * n}, SI _{M * 2n}Or SI _{M * n}, as follows:

SI _{2m * n}(i, j)=SI _{2m * 2n}(i, 2j), SI _{M * 2n}(i, j)=SI _{2m * 2n}(2i, j) or SI _{M * n}(i, j)=SI _{2m * 2n}(2i, 2j),

When carrying out right-shift operation, 2m * n convergent-divergent or m * 2n convergent-divergent moves to right 1 less with respect to 2m * 2n convergent-divergent, and m * n convergent-divergent moves to right 2 less with respect to 2m * 2n convergent-divergent.

4. multiple dimensioned compatible processing device in image and the video coding, comprise convergent-divergent and the forward processor that quantizes to combine, described convergent-divergent comprises input unit (210), multiple dimensioned compatible positive-going transition device (220) and the output device (240) that is connected successively with the forward processor that quantizes to combine, and it is characterized in that: described multiple dimensioned compatible scalar quantization device (230) is arranged between multiple dimensioned compatible positive-going transition device (220) and the output device (240);

Described multiple dimensioned compatible positive-going transition device (220) carries out multiple dimensioned compatible positive-going transition to 2m * 2n, 2m * n, m * 2n or m * n input block to be handled: f _{2m * 2n}=T _2m* F _{2m * 2n}* T _2n, f _{2m * n}=T _2m* F _{2m * n}* T _n, f _{M * 2n}=T _m* F _{M * 2n}* T _2nOr f _{M * n}=T _m* F _{M * n}* T _n, wherein, F _{2m * 2n}, F _{2m * n}, F _{M * 2n}Or F _{M * n}Represent the multiple dimensioned compatible positive-going transition input data of 2m * 2n, 2m * n, m * 2n or m * n respectively, f _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}Represent the multiple dimensioned compatible positive-going transition result of 2m * 2n, 2m * n, m * 2n or m * n respectively, T _2m, T _m, T _2nAnd T _nBe respectively 2m * 2m, m * m, 2n * 2n and n * n transformation matrix, and T _2mAnd T _m, T _2nAnd T _nSatisfy following relation:

$T_{2m}=\left[\begin{array}{cc}{T}_m& 0\\ 0& T_m^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_m& I_m\\ I_m& -I_m\end{array}\right],$ $T_{2n}=\left[\begin{array}{cc}{T}_n& 0\\ 0& T_n^{\′}\end{array}\right]\left[\begin{array}{cc}{I}_n& I_n\\ I_n& -I_n\end{array}\right],$

T ' wherein _mAnd T ' _nBe respectively m * m and n * n matrix, I _mAnd I _nIt is respectively m * m and n * n unit matrix; ALU (225) included in the described multiple dimensioned compatible positive-going transition device (220) is realized multiple dimensioned compatible positive-going transition processing with multiplying, add operation and shift operation;

Described multiple dimensioned compatible scalar quantization device (230) carries out multiple dimensioned compatible scalar quantization to 2m * 2n, 2m * n, m * 2n or m * n transform-domain signals to be handled: ${\mathrm{sf}}_{2m\×2n}=f_{2m\×2n}\⊗{\mathrm{QS}}_{2m\×2n},$ ${{\mathrm{qsf}}_{2m\×n}=f_{2m\×n}\⊗{\mathrm{QS}}_{2m\×n},\mathrm{qsf}}_{m\×2n}=f_{m\×2n}\⊗Q{S}_{m\×2n}$ Or $q{\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗{\mathrm{QS}}_{m\×n},$ Wherein, qsf _{2m * 2n}, qsf _{2m * n}, qsf _{M * 2n}Or qsf _{M * n}Represent the multiple dimensioned compatible scalar quantization result of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}, QS _{2m * n}, QS _{M * 2n}Or QS _{M * n}Represent the multiple dimensioned compatible scalar quantization coefficient matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}In (i, j) individual coefficient is QS _{2m * 2n}(i, and j)=1/ (Q (i, j) * (s _i* s _j) ²), s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, (i j) is the quantization step factor, QS to Q _{2m * n}(i, j)=2 * QS _{2m * 2n}(i, 2j), QS _{M * 2n}(i, j)=2 * QS _{2m * 2n}(2i, j), QS _{M * n}(i, j)=4 * QS _{2m * 2n}(2i, 2j); ALU (235) included in the described multiple dimensioned compatible scalar quantization device (230) is realized multiple dimensioned compatible scalar quantization processing with division arithmetic, multiplying, add operation and shift operation.

5. the processing method of multiple dimensioned compatible processing device in a kind of image according to claim 4 and the video coding is characterized in that:

Multiple dimensioned compatible positive-going transition device is exported the f that obtains _{2m * 2n}, f _{2m * n}, f _{M * 2n}Or f _{M * n}In one or several combination carry out multiple dimensioned compatible scalar quantization and handle, finish just reverse scalar quantization and handle, as follows:

${\mathrm{qsf}}_{2m\×2n}=f_{2m\×2n}\⊗{\mathrm{QS}}_{2m\×2n},{\mathrm{qsf}}_{2m\×n}=f_{2m\×n}\⊗{\mathrm{QS}}_{2m\×n},{\mathrm{qsf}}_{m\×2n}=f_{m\×2n}\⊗{\mathrm{QS}}_{m\×2n}$ Or

$q{\mathrm{sf}}_{m\×n}=f_{m\×n}\⊗{\mathrm{QS}}_{m\×n},$

Wherein, qsf _{2m * 2n}, qsf _{2m * n}, qsf _{M * 2n}Or qsf _{M * n}Represent the multiple dimensioned compatible scalar quantization result of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}, QS _{2m * n}, QS _{M * 2n}Or QS _{M * n}Represent the multiple dimensioned compatible scalar quantization coefficient matrix of 2m * 2n, 2m * n, m * 2n or m * n respectively, QS _{2m * 2n}In (i, j) individual coefficient is QS2 _{M * 2n}(i, and j)=1/ (Q (i, j) * (s _i* s _j) ²), s wherein _iBe the mould of the capable vector of i in 2m * 2m transformation matrix, s _jBe the mould of the capable vector of j in 2n * 2n transformation matrix, (i j) is the quantization step factor, QS to Q _{2m * n}(i, j)=2 * QS _{2m * 2n}(i, 2j), QS _{M * 2n}(i, j)=2 * QS _{2m * 2n}(2i, j), QS _{M * n}(i, j)=4 * QS _{2m * 2n}(2i, 2j).

6. multiple dimensioned compatibility processing method in a kind of image according to claim 5 and the video coding is characterized in that: utilize 2m * 2n integer scalar quantization coefficient matrix QSI _{2m * 2n}Structure 2m * n, m * 2n or m * n integer scalar quantization coefficient matrix QSI _{2m * n}, QSI _{M * 2n}Or QSI _{M * n}, as follows:

QSI _{2m * n}(i, j)=QSI _{2m * 2n}(i, 2j), QSI _{M * 2n}(i, j)=QSI _{2m * 2n}(2i, j) or QSI _{M * n}(i, j)=QSI _{2m * 2n}(2i, 2j),

When carrying out the right-shift operation under the same degree quantification, 2m * n scalar quantization or m * 2n scalar quantization moves to right 1 less with respect to 2m * 2n scalar quantization, and m * n scalar quantization moves to right 2 less with respect to 2m * 2n scalar quantization.