CN101325714B - Method and apparatus for processing transformation data, method and apparatus for encoding and decoding - Google Patents

Abstract

The embodiment of the invention claims a processing method for transforming data, comprising: estimating and analyzing the numerical value range of the image data after two transformations based on pre-setting two needed transforming matrixes needed by two transformations; estimating the characteristic different value of numerical value of two transformations based on two transformed numerical value; applying the first transformation of two transformations to the data to be transformed, and compensating the data after the first transformation based on the different value of estimated numerical value. The embodiment of this invention further discloses a processing device for transforming data, method and device of coding and decoding. Appling the embodiment of this invention, which can adjust the numerical value of the transformed data so as to enable the numerical values of the image data to be the same after different transformation and really reflect the influence of the transformation to the data when applying the self-adoptive block transformation technique, and to select the transformation with better effect and improve the coding efficiency.

Description

Transform data processing method and device and coding, coding/decoding method and device

Technical field

The present invention relates to technology of video compressing encoding, particularly a kind of transform data processing method and device, and a kind of coding, coding/decoding method and device and system.

Background technology

For reducing the data volume of video data when transmitting or store, generally need carry out compressed encoding to video data.In the video compression coding field, conversion is an important technology, and its effect is that picture material and information in a sub-picture, the zone are concentrated on a certain specific region after conversion, so that video compression algorithm carries out more efficient compression to this part content.Then to after the conversion data being quantized, promptly form after the processing such as entropy coding video data behind the compressed encoding.

In the standard of coding and decoding video, as MPEG-2, H.264, AVS has used converter technique.In these standards, a zone in piece image or the image is divided into plurality of small blocks or subregion, is called sub-piece, and conversion is that unit carries out with sub-piece.Generally speaking, the size of sub-piece can be 4x4 or 8x8, and wherein, 4 and 8 all is unit with the image pixel.

One section video file is made up of several video images, and piece image comprised abundant content usually, and the characteristic that different piece had of image is also different.Therefore, if divide with the sub-piece of same size (as the 8x8 size) with all images in one section video or with piece image, then it is carried out conversion, its effect differs and reaches optimum surely, promptly can not effectively the content in all sub-pieces effectively be concentrated on a certain specific region after conversion.Based on this, the adaptive block transformation technology has been proposed, its principle is: a specific region is divided according to different sub-block sizes respectively, sub-piece at different sizes (for example carries out different conversion then, image is divided into the piece of 4x4 and 8x8 respectively, the 4x4 piece is used the 4x4 conversion, the 8x8 piece is used the 8x8 conversion), judge according to certain criterion which kind of conversion can more effective information with piece concentrate on the specific region under different change situation afterwards.At last more excellent transformation results is stored.Decoding end is when decoding to the image after the conversion of process aforesaid way, obtain the information of change of scale (as 4x4 or 8x8) according to corresponding information in the code stream, use corresponding inverse transformation (as 4x4 inverse transformation or 8x8 inverse transformation) that this zone is handled then, obtain original video data.

The concrete mode that above-mentioned adaptive block transformation technology is used in H.264 is, definition one cover 4x4 transformation matrix and 8x8 transformation matrix, and characteristics according to its 4x4 conversion and 8x8 conversion, formulated the quantization table of a cover encoding and decoding end respectively, with data respectively through the 4x4 conversion and the result after quantizing with through the 8x8 conversion and the result after quantizing compare definite more excellent mapping mode.Wherein, because conversion has similar transform characteristics with 8x8 in the 4x4 conversion H.264, the quantization table of supporting formulation separately in addition can guarantee that the 4x4 piece is through the 4x4 conversion and value data scope after quantizing and the value data scope basically identical of 8x8 piece after 8x8 conversion and quantification.Therefore, the adaptive block transformation technology in H.264 can effectively improve code efficiency.

But, in compression encoding process to video, may need to merge different conversion for various purposes, for example, the 4x4 transformation matrix is based on DCT's, the 8x8 conversion then is based on small echo, and this two covers transformation matrix does not have too many identical transform characteristics probably, and identical data cause through the number range change degree varies of data after these conversion.Because quantification can cause data message impaired, no matter in quantizing process, use identical or different quantization table to quantize, cause because of identical data its number range after different conversion changes degree varies, then can cause conversion data afterwards inconsistent through the extent of damage after quantizing.In this case, can't use certain judgment criterion to determine more excellent mapping mode, thereby can't effectively improve digital coding efficient.

Summary of the invention

In view of this, the embodiment of the invention provides a kind of method and apparatus of adjusting the coded data number range, so that the number range of data after different conversion is consistent substantially.Correspondingly, also provide a kind of coding method, a kind of coding/decoding method and device and system, will adjust parameter at coding side and write in the code stream, can adjust accordingly the data that receive according to the adjustment parameter that receives in decoding end.

For achieving the above object, the embodiment of the invention adopts following technical scheme:

A kind of transform data processing method comprises:

According to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion;

According to the number range after described two kinds of conversion, estimate the number range feature difference of two kinds of conversion, the number range feature difference of described two kinds of conversion be view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, or view data is respectively through the difference of number range after two kinds of conversion;

Receive data to be transformed, and these data are used first kind of conversion in described two kinds of conversion, and according to the number range feature difference of estimating compensate after this first kind of conversion data so that the number range behind the number range behind first kind of transform and quantization and the second kind of transform and quantization be consistent.

Above-mentioned first kind is transformed to a kind of in described two kinds of conversion.In like manner, another conversion in described two kinds of conversion can be referred to as second kind of conversion.Described two kinds of conversion also can be referred to as first conversion and second conversion respectively, or conversion A and conversion B.

A kind of transform data processing unit comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit,

The described first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer described number range difference unit;

Described second value scope estimation unit is used for the required transformation matrix of second conversion according to default two kinds of conversion, the number range of estimated image data after through second conversion, and offer described number range difference unit;

Described number range difference unit, be used for according to the number range of passing through respectively after first, second conversion, estimate the number range feature difference of two kinds of conversion, and offer described conversion compensating unit, the number range feature difference of described two kinds of conversion be view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, or view data is respectively through the difference of number range after two kinds of conversion;

Described conversion compensating unit, be used to receive data to be transformed, and these data are used first kind of conversion in described two kinds of conversion, and according to the described number range feature difference of estimating compensate after this first kind of conversion data so that the number range behind the number range behind first kind of transform and quantization and the second kind of transform and quantization be consistent.

A kind of coding method comprises:

Receive data to be transformed;

Described data to be transformed are carried out first conversion, obtain the data after first conversion;

Described data to be transformed are carried out second conversion, obtain the data after second conversion;

Determine to adjust parameter according to the parameter of second conversion, data after first conversion and the data after second conversion, according to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion;

Described adjustment parameter is write in the encoding code stream.

A kind of coding/decoding method comprises:

Receive code stream, described code stream is decoded, obtain the data after first conversion and adjust parameter;

According to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion.

A kind of code device comprises:

The Data Receiving unit is used to receive data to be transformed;

Converter unit is used for the data described to be transformed that receiving element receives are carried out first conversion, obtains the data after first conversion; Described data to be transformed are carried out second conversion, obtain the data after second conversion;

First adjustment unit is determined to adjust parameter according to the parameter of second conversion, data after first conversion and the data after second conversion, according to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion;

R/w cell is used for described adjustment parameter is write encoding code stream.

A kind of decoding device comprises:

The code stream receiving element is used to receive code stream;

Decoding unit is decoded to described code stream, obtains the data after first conversion and adjusts parameter;

Second adjustment unit is used for according to the data after parameter adjustment first conversion of the described adjustment parameter and second conversion.

A kind of coding/decoding system comprises: encoding apparatus and decoding apparatus;

Described code device comprises:

The Data Receiving unit is used to receive data to be transformed; Converter unit is used for the data described to be transformed that receiving element receives are carried out first conversion, obtains the data after first conversion; Described data to be transformed are carried out second conversion, obtain the data after second conversion;

First adjustment unit is determined to adjust parameter according to the parameter of second conversion, data after first conversion and the data after second conversion, according to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion; R/w cell is used for described adjustment parameter is write encoding code stream.

Described decoding device comprises:

The code stream receiving element is used to receive code stream; Decoding unit is decoded to described code stream, obtains the data after first conversion and adjusts parameter; Second adjustment unit is used for according to the data after parameter adjustment first conversion of the described adjustment parameter and second conversion.

A kind of coding method comprises:

Receive data to be transformed and coding parameter information;

Described data to be transformed are carried out first conversion, obtain the data after first conversion;

Determine to adjust parameter according to the parameter of the coding parameter information and second conversion, according to the data after described first conversion of the parameter adjustment of described adjustment parameter and described second conversion;

Described adjustment parameter is write in the encoding code stream.

A kind of code device comprises:

The 3rd receiving element is used to receive data to be transformed and coding parameter information;

The 3rd converter unit is used for the data to be transformed that described the 3rd receiving element receives are carried out first conversion, obtains the data after first conversion;

The 3rd adjustment unit, be used for the information of coding parameter that receives according to the 3rd receiving element and the parameter of second conversion and determine to adjust parameter, the data after first conversion that obtains according to described the 3rd converter unit of the parameter adjustment of described adjustment parameter and described second conversion;

The 3rd r/w cell is used for the adjustment parameter that described the 3rd adjustment unit obtains is write encoding code stream.

As seen from the above technical solution, in embodiments of the present invention,, estimate that video data is respectively through the number range after two kinds of conversion at first according to default two kinds of transformation matrixs that conversion is required; Then, according to the number range after described two kinds of conversion and two kinds of conversion corresponding quantitative point separately, estimate that video data is respectively through two kinds of conversion and the corresponding difference that quantizes the back number range.At last, after wherein a kind of conversion of data application to be transformed, the difference of the number range that utilization is estimated compensates the data after this first kind of conversion.Since when carrying out the compensation of number range according to the relation of number range behind two kinds of transform and quantizations, therefore, can realize to make the number range behind the transform and quantization consistent after compensating with number range behind the another kind of transform and quantization, thereby when application self-adapting piece converter technique, can truly reflect the influence of transfer pair data, select the more excellent conversion of effect, and then improve code efficiency.Decoding method provided by the invention and device, can guarantee that data area after first conversion and the data area after second conversion are consistent, to adjust parameter simultaneously and write in the code stream, can the data that receive be handled accordingly according to adjusting parameter in decoding end like this.

Description of drawings

Fig. 1 is the overview flow chart of transform data processing method in the embodiment of the invention;

Fig. 2 is the overall construction drawing of transform data processing unit in the embodiment of the invention;

Fig. 3 is the particular flow sheet of transform data processing method in the embodiment of the invention one;

Fig. 4 is the concrete structure figure of transform data processing unit in the embodiment of the invention one;

Fig. 5 is a transform data processing method particular flow sheet in the embodiment of the invention two;

Fig. 6 is a transform data processing method particular flow sheet in the embodiment of the invention three;

Fig. 7 is the concrete structure figure of transform data processing unit in the embodiment of the invention three;

Fig. 8 is a transform data processing method particular flow sheet in the embodiment of the invention four;

Fig. 9 is the concrete structure figure of transform data processing unit in the embodiment of the invention four;

The coding method schematic flow sheet that Figure 10 provides for the embodiment of the invention five;

The coding/decoding method schematic flow sheet that Figure 11 provides for the embodiment of the invention six;

The code device structural representation that Figure 12 provides for the embodiment of the invention seven;

Figure 13 is the decoding device structural representation that the embodiment of the invention eight provides;

Figure 14 is the coding method schematic flow sheet that the embodiment of the invention nine provides;

Figure 15 is the code device structural representation that the embodiment of the invention ten provides.

Embodiment

For the purpose, technological means and the advantage that make the embodiment of the invention is clearer, the embodiment of the invention is described in further detail below in conjunction with accompanying drawing.

In embodiments of the present invention, at first analyze the variation of data value data scope after two kinds of different conversion, then according to the variation of value data scope behind the different transform and quantizations, data after a kind of conversion are wherein compensated, to guarantee that identical data to be encoded are through the number range maintenance basically identical of the number range of data behind this transform and quantization with data behind another kind of transform and quantization, to improve code efficiency.

Fig. 1 is the overview flow chart of transform data processing method in the embodiment of the invention.As shown in Figure 1, this method comprises:

Step 101, according to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion.

Step 102 according to the number range after described two kinds of conversion, is estimated the number range feature difference of two kinds of conversion.

According to different compensation ways, the number range feature difference value of above-mentioned two kinds of conversion is also different, can for view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, perhaps also can for view data respectively through the difference of number range after two kinds of conversion.When this feature difference is during respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, when calculating this feature difference, the quantization step that will further answer also according to two kinds of transfer pairs.

Step 103 is used a kind of in described two kinds of conversion to data to be transformed, and compensates data after this conversion according to the difference of the number range of estimating.

Fig. 2 is the overall construction drawing of transform data processing unit in the embodiment of the invention.As shown in Figure 2, this device comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit.

Wherein, the first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer the number range difference unit.

Second value scope estimation unit is used for the required transformation matrix of second conversion according to default two kinds of conversion, the number range of estimated image data after through second conversion, and offer the number range difference unit.

The number range difference unit is used for estimating the number range feature difference of two kinds of conversion, and offering the conversion compensating unit according to the number range of passing through respectively after first, second conversion.

The conversion compensating unit is used for data to be transformed are used first kind of conversion, and compensates data after this first kind of conversion according to the described number range feature difference of estimating.

In said method and device, the mode of data can be after the compensation conversion: according to the relation between the number range of data after two kinds of conversion, determine a suitable adjustment factor, to multiply by this adjustment factor through a kind of data of conversion gained wherein, quantize according to its corresponding quantitative point again, thereby make the number range of passing through this conversion and quantizing back gained data with consistent through the number range of gained data behind another transform and quantization.

Perhaps, the mode of the data after the compensation conversion can also be: according to the relation between the number range of data after two kinds of conversion, the point of quantification that a kind of transfer pair is wherein answered is adjusted, according to adjusted point of quantification these conversion gained data are quantized again, thereby also can realize making the number range of this conversion and quantification back gained data with consistent through the number range of gained data behind another transform and quantization.

Or, the mode of the data after the compensation conversion can also be: according to the relation between the number range of data after two kinds of conversion, for a kind of conversion wherein rebulids quantization table, according to the quantization table of rebuilding these conversion gained data are quantized again, thereby can realize making the number range of this conversion and quantification back gained data equally with consistent through the number range of gained data behind another transform and quantization.

Below, the specific embodiment of the present invention under above-mentioned three kinds of different disposal modes is described by different embodiment.Wherein, embodiment one～two is at first kind of processing mode, and embodiment three is at second kind of processing mode, and embodiment four is at the third processing mode.

Embodiment one:

In the present embodiment, supposing has coded data block C, and conversion A and conversion B.The yardstick of conversion A is n * n, and the yardstick of conversion B is m * m.When being encoded, uses data block C point of quantification QP _s, conversion A and conversion B are QP in point of quantification _sThe time quantization step that found be respectively QTAB1[s] and QTAB2[s].Here, when data were quantized, the corresponding point of quantification of each different coding region (for example, data block C) found the corresponding quantitative step-length according to this point of quantification, carries out quantification treatment.Adjustment is through the number range of conversion A and corresponding quantification gained data, makes it consistent with the number range of gained data after conversion B and the corresponding quantification.In the present embodiment, the number range feature difference of two kinds of conversion is that view data is respectively through two kinds of conversion and the corresponding difference that quantizes the back number range.

Fig. 3 is the particular flow sheet of transform data processing method in the embodiment of the invention one.As shown in Figure 3, this method comprises:

Step 301, according to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion.

In this step, encoding block C is divided into sub-piece according to the yardstick of conversion A, B, conversion A, B are applied to respectively on each sub-piece.According to the transformation matrix of conversion A, B correspondence, the number range of each pixel data after conversion in the sub-piece of estimating to divide.Under any one conversion, estimate that the mode of number range is identical after the conversion, to estimate that behind conversion A the number range of each pixel data is an example in the sub-piece, concrete method of estimation is described below.

Suppose that encoding block C size is MxN, T is the transformation matrix of conversion A correspondence, and T ' is the transposed matrix of T, then uses conversion A for encoding block C and is specially: calculate

Wherein, S is the zoom factor matrix, and this matrix is used to make [TCT '] data normalization afterwards, is the matrix multiple symbol,

Each corresponding element of representing matrix multiplies each other.According to matrix multiplication can calculate the number range increment of [TCT '].

At first estimate [TCT '] number range increment with respect to C: j row coefficient absolute value sum is d among the T ' _j, then i is capable among these row and the C obtains the number range of data after multiplying each other, and maximum becomes among the C the capable j columns of i according to the d of number range _jDoubly.Because computer and hardware processor all are that unit carries out information stores with the binary system, thus represent number range with binary form herein, will be listed as here and C in i is capable multiplies each other afterwards that the number range maximal increment of data is expressed as log ₂(d _j).As above can calculate the greatest measure scope increment of each element among the C of [TCT '] back.

Estimate then

Number range increment with respect to [TCT ']: (k, the absolute value of element h) are S if be positioned among the S _{(k, h)}, then

Afterwards, with respect to [TCT '], (k, number range increment h) is log to point ₂(S _{(k, h)}).

With the number range increment addition that above-mentioned two steps are estimated, can estimate to draw

The greatest measure scope increment of every bit among the C afterwards.Again according to the number range of encoding block C itself, promptly can obtain conversion after, the numerical value greatest measure scope of each point in each sub-piece.

What estimation obtained through aforesaid way is the greatest measure scope of encoding block C after conversion, and in actual coding a common encoding block C, because the data among the encoding block C are obeyed certain mathematical distribution, and the very difficult greatest measure scope that all reaches self, therefore the number range after conversion does not reach the greatest measure scope probably.Therefore, preferably, when carrying out the number range estimation, further consider the mathematical distribution situation of encoding block C.

Particularly, suppose that the numerical value among the piece C holds certain mathematical distribution, conversion A is applied in this distribution, can calculate the most possible zone that occurs of number range of C data after conversion again by said method.For example, the mathematical distribution model of supposing the data of MxN size data piece C be P (x, y), the capable each point maximum possible of this data block i numerical value scope is expressed as f (P (x, i)), (0＜=x＜=M), wherein f () is the mapping relations of trying to achieve according to mathematical distribution model P, its implication is point (x, y), (0＜=x＜=M, the number range that the pairing maximum possible of 0＜=y＜=N) occurs, promptly (x, y) most possible numerical value is 2 to point when coding ^{F (P (x, y))}, with 2 ^{F (P (x, y))}Be made as v _iSimultaneously, the coefficient value of supposing conversion T ' j row each point be T ' (j, y), (0＜=y＜=N), then i is capable among T ' j row and the C multiply each other after, C i line data number range maximal increment is:

Can calculate thus through conversion

The number range that back data block C each point maximum possible occurs.

After the process aforesaid way carries out the estimation of number range, can obtain the number range of sub-piece each point data behind conversion A among the encoding block C:

A11?A12.........A1(n-1)A1n

A21?A22.........A2(n-1)A2n

An1?An2..........An(n-1)Ann

And the number range of sub-piece each point data behind conversion B among the encoding block C is:

B11?B12?B13................?B1(m-2)B1(m-1)B1mm

B21?B22?B23.................B2(m-2)B2(m-1)B2mm

B31?B32?B33.................B3(m-2)B3(m-1)B3mm

Bm1?Bm2?Bm3..............Bm(m-2)Bm(m-1)Bmm

Step 302, at two kinds of conversion, the mean values scope behind the difference computational transformation.

In this step, identical for the mean values range computation method of two kinds of conversion.Be example still with conversion A, the mode of the mean values scope behind the computational transformation can for:

Wherein, Aij is the number range of the interior capable j row of the i pixel data of sub-piece behind the conversion A that obtains in the step 301.Similarly, the mean values range computation mode behind the conversion B is:

Wherein, Bij is the number range of the interior capable j row of the i pixel data of sub-piece behind the conversion B.

Step 303 according to mean values scope after the conversion that obtains in the step 302, is calculated the number range difference after two kinds of conversion and the corresponding quantification.

In this step, at first calculate the number range after two kinds of conversion and corresponding quantification.Still turn to the example explanation with conversion A and respective amount.As previously mentioned, encoding block C is used point of quantification QP _s, corresponding to this point of quantification, the quantization step that conversion A finds is QTAB1[s], utilize the Avr that calculates in the step 202 _AExpression is the average number range of gained data block each point behind conversion A, and number range represents that with binary form then coded data gained digital average value behind conversion A is

Usually in coding and decoding video the quantizing process of data can be expressed as (X * QTAB[s])＞＞shift[s] (1), wherein X is a numerical value to be quantified, s is a point of quantification, QTAB[n] be the quantization step that in quantization table, checks in according to point of quantification n, shift[s] be the pairing displacement of point of quantification s.According to above-mentioned steps as can be known, utilize quantization step QTAB1[s] quantize back gained digital average value and be

The number range of representing these data with binary form is (Avr _A+ log ₂QTAB1[s])＞＞shift[s].In like manner, through conversion B and with QTAB2[s] be the quantification of quantization step after, the number range of gained data is (Avr _B+ log ₂QTAB2[s])＞＞shift[s].

Next, the number range difference after two kinds of conversion of calculating and the corresponding quantification.Concrete account form is: calculates that the multiple between numerical value concerns after two kinds of conversion and the corresponding quantification, promptly

The difference that this multiple relation is converted between the number range that binary form represents is:

D '=(Avr just _B+ log ₂QTAB2[s])-(Avr _A+ log ₂QTAB1[s]).

Step 304 according to the difference in the step 303, is determined the adjustment factor of conversion A.

In this step, according to the number range difference that obtains in the step 303, the adjustment factor of definite conversion A is q=2 ^{D '}

Desired data is priori data in the above-mentioned steps computational process, i.e. data that can obtain before coding, so above-mentioned steps can be finished before coding and the result is preserved, and need not to calculate in cataloged procedure.

The adjustment factor of conversion A may adjust because of the difference of coded data in actual coding, so the adjustment factor of conversion A can be done suitable adjustment according to the characteristic of coded data when coding.Described coded data characteristic can be the actual average number range of data block after conversion A.And the adjustment factor of conversion A write in the encoding code stream.The adjustment factor of conversion A can write in the sequence head or image head or slice header or macro block head of code stream.The occupation mode of the adjustment factor of conversion A is identical with said method, repeats no more here.

Step 305 is used conversion A to encoding block C, obtains the data after the conversion, and the gained data all be multiply by the adjustment factor of conversion A.

In this step, encoding block C is used conversion A, its detailed process is identical with existing mode, just repeats no more here.The data that conversion is obtained multiply by the adjustment factor in the step 204, and promptly each data for gained after the conversion have: C _Ij'=C _Ij2 ^{D '}, wherein, C _IjBe the pixel data value of the capable j row of i after the conversion, C _Ij' be the reconstructed value of adjusting the pixel data of the capable j row of back i.

Step 306 utilizes quantization step that conversion A finds to each C _Ij' quantize.

According to existing mode to conversion A after each C through adjusting _Ij' quantize.Adjust each C of factor pair owing to utilize in the step 205 _IjCarried out rebuilding and obtained C _Ij', and this adjustment factor is to determine that according to the difference of number range after two kinds of conversion and the corresponding quantification therefore, this step is to each C _Ij' the number range that obtains data after quantizing just can be consistent with the number range that encoding block C obtains data after through conversion B and corresponding quantification.

Concrete derivation is as follows: because the displacement in the quantizing process is identical with conversion B to influence to conversion A, below just no longer consider the influence of displacement.

As previously mentioned, the mean values scope behind the encoding block C application conversion A is Avr _A, then coded data institute's value behind conversion A is Multiply by the result who adjusts after the factor so is Number range with binary representation is Avr _B+ log ₂QTAB2[s]-log ₂QTAB1[s]; Utilizing quantization step QTAB1[s again] result that obtains after quantizing is Number range with binary representation is Avr _B+ log ₂QTAB2[s].And the mean values scope behind the encoding block C application conversion B is Avr _B, then coded data institute's value behind conversion A is

Utilizing quantization step QTAB2[s again] result that obtains after quantizing is

Number range with binary representation is Avr _B+ log ₂QTAB2[s].As seen, after the operation of same encoding block C through step 205 and 206, with identical through the number range of gained data after conversion B and the corresponding quantification.Conversion A and the consistent purpose of the final number range of conversion B have been realized making.

What so far, present embodiment provided finishes the transform data process flow.

Handle according to the encode transform data of side of said method, thereby select more excellent mapping mode among conversion A and the conversion B.If the final mapping mode of selecting is conversion A, the data that then send to decoding end are the data after conversion A and corresponding quantification and entropy coding etc. are handled.At this moment, after decoding end is carried out inverse quantization to data according to conversion A corresponding quantitative point, need carry out inverse transformation again after divided by the adjustment factor of conversion A, to rebuild the coding side data to the result behind the inverse quantization.Wherein, the adjustment factor of conversion A can obtain in decoding end according to the process of above-mentioned steps 301～304.If decoding end for avoiding using division, can allow decoded data reach decoded data divided by the purpose of adjusting the factor by multiplication and displacement mode.

If encoding code stream has comprised the adjustment factor of conversion A when coding, then decoding end includes the place of adjusting the factor from code stream, as sequence head, image head, slice header, macro block head, obtains the adjustment factor of conversion A.The occupation mode of the adjustment factor of conversion A is identical with said method, repeats no more here.

Present embodiment also provides a kind of embodiment of transform data processing unit, can be used to implement above-mentioned method shown in Figure 3.Fig. 4 is the concrete structure figure of this execution mode down conversion data processing equipment.As shown in Figure 4, this device comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit.Wherein, the conversion compensating unit comprises that adjusting the factor determines subelement, varitron unit, conversion reprocessing subelement and quantize subelement.

The first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer the number range difference unit.

Second value scope estimation unit is used for the required transformation matrix of second conversion according to default two kinds of conversion, the number range of estimated image data after through second conversion, and offer the number range difference unit.

The number range difference unit, be used for according to the number range and the corresponding separately point of quantification of first, second conversion of passing through respectively after first, second conversion, the estimated image data quantize the difference of back number range through first, second conversion and correspondence respectively, and offer the conversion compensating unit.

In the conversion compensating unit, to adjust the factor and determine subelement, the difference that the number range that provides according to described number range difference unit is provided is determined to adjust the factor, and offers conversion reprocessing subelement.The varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered conversion reprocessing subelement.Conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment factor, and the result is offered the quantification subelement.Quantize subelement, the result who multiply by after the adjustment factor is quantized according to view data corresponding quantitative point.

If coding side will be adjusted the factor and write in the code stream, then also need increase accordingly and adjust factor writing unit, be used for the adjustment factor is write code stream.

In the present embodiment, at first the data behind the conversion A be multiply by one and adjust the factor, and then carry out quantification treatment.By the quantizing process of formula (1) expression as can be seen, when quantification treatment, at first to multiply each other, and then multiplied result is done dextroposition, just give up low level information wherein with quantization step.Multiply by this adjustment factor and this result be multiply by quantization step QTAB1[s through the data behind the conversion A so] after, possible low level is not 0, but has still concentrated partial information, at this moment when being shifted, will cause the loss of data precision.Based on this, for reducing the loss of data precision in the quantizing process, a kind of evolution processing mode has been proposed also on the present embodiment basis, also increase the number range that the mode that is shifted changes data behind conversion A by the adjustment factor that multiply by conversion A.

Particularly, flow process corresponding shown in Figure 3, the operation of step 301～303 is constant, and in step 304, the adjustment factor of the conversion A that determines becomes q '=2 ^{D '}* 2 ⁿ, be referred to as the adjustment coefficient, wherein, the value of n is decided according to coding side hardware store scope, and after the data after guaranteeing conversion A multiplied each other with adjustment factor q ', the data of its subsequent operation did not all exceed under the condition of hardware store scope, and the value of n is the bigger the better.Information in the data behind the conversion A can be concentrated on a high position as far as possible like this.

Next, in step 305, conversion gained data and above-mentioned adjustment factor q ' are multiplied each other.

At last, when in step 306 data being quantized, the displacement figure place of point of quantification correspondence is increased the n position again, just after executing quantization operation, n position again moves to right quantized result.Like this, can will make 2 of number range expansion by the change because of the adjustment factor in the top amended step 304 ⁿDoubly compensation is returned.

Say that simply the operation of above-mentioned amended step 305～306 is: (2 ^{D '}* 2 ⁿ* X * QTAB[s])＞＞shift[s]+n, wherein, X represents the data after the conversion.See the operation (2 in this operation and former step 305～306 intuitively ^{D '}* X * QTAB[s])＞＞shift[s] to compare, it comes to the same thing, and when still realizing under hardware platform, the former loss of significance is littler than the latter.This be because, in amended step 305, multiply by adjust factor q ' after, data X is extended 2 ^{D '}* 2 ⁿDoubly, with respect to the data after former step 305 operation, its useful information has been exaggerated, and concentrates on a high position more, quantizes in amended step 306 like this in the related shifting process, is shifted and the useful information that loses has reduced relatively.Like this, just, can more effectively improve data precision.

Adopt the evolution mode of the foregoing description one when coding side after, when decoding end is carried out inverse transformation and inverse quantization for the data of handling through conversion A, also to carry out corresponding inverse operation.Particularly, when inverse quantization, the figure place that will be shifted reduces n, and before the inverse transformation, with the adjustment factor q ' of inverse quantization result divided by conversion A, carries out inverse transformation then, to rebuild the coding side data more accurately behind inverse quantization.

Certainly, corresponding above-mentioned evolution mode the present invention also provides corresponding evolution apparatus structure, this structure and similar shown in Figure 4, difference are that this conversion compensating unit comprises adjustment coefficient and displacement side-play amount subelement, varitron unit, conversion reprocessing subelement and quantizes subelement.All the device with shown in Figure 4 is identical with function and for the structure of the first number range estimation unit, second value scope estimation unit and number range difference unit.

In the conversion compensation subelement of this evolution device, adjust coefficient and displacement side-play amount subelement, displacement side-play amount when being used for according to the number range feature difference of estimating definite adjustment coefficient and quantification, and offer conversion reprocessing subelement and quantification subelement respectively.The varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered conversion reprocessing subelement.Conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment coefficient, and the result is offered the quantification subelement.Quantize subelement, the quantization step of answering according to described displacement side-play amount and first kind of transfer pair quantizes multiply by the result who adjusts after the factor.

In the above-described embodiment, when calculating the difference of number range after two kinds of conversion and the corresponding quantification, utilization be two kinds of mean values scopes after the conversion.More accurate for number range is estimated, can also adopt the mode of embodiment two to carry out.

Embodiment two:

In the present embodiment, similar with embodiment one, the number range feature difference of two kinds of conversion be view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, but different among the concrete account form of this difference and the embodiment one.

Fig. 5 for provide in the present embodiment to transform data processing method particular flow sheet.As shown in Figure 5, this method comprises:

Step 501, according to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion.

The estimation mode of this step is identical with step 301 among the embodiment one, just repeats no more here.

Step 502, the mean values scope behind the computational transformation B.

Identical among the mode of mean values scope and the embodiment one behind the computational transformation B in this step, be specially

Wherein, Bij is the number range of the interior capable j row of the i pixel data of sub-piece behind the conversion B.

Step 503 according to the mean values scope behind conversion B that number range and the step 502 through each data behind the conversion A that obtain in the step 501 obtain, is calculated the number range difference after two kinds of conversion and the corresponding quantification.

In this step, identical among the method principle the during difference of evaluation scope and the embodiment one, difference is, the difference of the data area in the present embodiment is calculated according to the number range of each data behind the conversion A and the mean values scope behind conversion B.Particularly, be divided into sub-piece for encoding block C according to the yardstick of conversion A after, the data area difference of the pixel data of the capable j row of i behind corresponding and conversion B behind the conversion A is in this sub-piece: d ' _{(i, j)}=(Avr _B+ log ₂QTAB2[n])-(Aij+log ₂QTAB1[n]), (wherein, Aij is the number range of the capable j row of i pixel data in the sub-piece behind conversion A for 1＜=i＜=m, 1＜=j＜=m).

Step 504 according to the number range difference that obtains in the step 503, is determined the adjustment factor of conversion A.

In this step, mode that determine to adjust the factor is identical with embodiment one, and the pixel data that is listed as at the capable j of i in the sub-piece and deciding, and is specially

The adjustment factor of trying to achieve like this, is that the variation according to location of pixels changes.

Step 505 is used conversion A to encoding block C, obtains the data after the conversion, and the equal correspondence of gained data be multiply by the adjustment factor of conversion A.

In this step, when multiplying each other, with result after the pixel data conversion of the capable j row of i and the corresponding factor q that adjusts with the adjustment factor _IjMultiply each other, promptly

Like this, will make that the adjustment of transformation results is more accurate.

Step 506 utilizes quantization step that conversion A finds to each C _Ij' quantize.

So far, promptly finished the compensation to the gained data of conversion A, its compensation principle is identical with the derivation of embodiment one, just repeats no more here.Its difference is, in compensation process, the calculating of adjusting the factor changes at the variation of pixel data position, make that the result of compensation is more accurate, make the consistent degree of the number range of gained data after final two kinds of conversion and the corresponding quantification higher, thereby can select more excellent conversion more effectively, further improve code efficiency.

The adjusting the factor according to the pixel position and may adjust because of the difference of coded data of conversion A in actual coding is so the adjustment factor of conversion A can be done suitable adjustment according to the characteristic of coded data when encoding.Described coded data characteristic can be the numerical value of each location of pixels of data block after conversion A.And the adjustment factor of conversion A write in the encoding code stream.The adjustment factor of conversion A can write in the sequence head or image head or slice header or macro block head of code stream.The occupation mode of the adjustment factor of conversion A is identical with said method, repeats no more here.

Method in the present embodiment also can adopt device shown in Figure 4 among the embodiment one to realize.

Consider the problem of compensation accuracy, present embodiment also provides a kind of compensation way of evolution, is specially: according to the difference of number range after two kinds of conversion that obtain and the corresponding quantification, determine to adjust factor f _{(i, j)}With coefficient side-play amount s _{(i, j)}, when compensating, at first utilize f _{(i, j)}Multiply each other with the data after the conversion, and then this result of product is added coefficient side-play amount s _{(i, j)}, purpose is to allow following relation set up f _{(i, j)}* Aij+s _{(i, j)}=Avr _B, (1＜=i＜=n, 1＜=j＜=n), thus the accuracy of assurance compensation.At last will add that again the result after the side-play amount quantizes according to conversion A corresponding quantitative step-length.Correspondingly, in decoding end, with coding side relative set f _{(i, j)}And s _{(i, j)}, and after the data for the treatment of inverse quantization that receive are carried out inverse quantization, deduct s _{(i, j)}Again divided by f _{(i, j)}, and then carry out inverse transformation, with accurate reconstruction coding side view data.

Compensation way for above-mentioned evolution, the present invention also provides corresponding evolution apparatus structure, this structure and similar shown in Figure 4, difference are that conversion compensating unit wherein comprises to be adjusted the factor and side-play amount subelement, varitron unit, conversion reprocessing subelement and quantize subelement.All the device with shown in Figure 4 is identical for the 26S Proteasome Structure and Function of other first number range estimation unit, second value scope estimation unit, number range difference unit.

In the conversion compensating unit, adjust the factor and coefficient side-play amount subelement, the number range difference that is used for providing according to described number range difference unit is determined to adjust the factor and coefficient side-play amount, and offers conversion reprocessing subelement.The varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered conversion reprocessing subelement.Conversion reprocessing subelement is used for that the data after first kind of conversion be multiply by definite adjustment factor and adds side-play amount, and the result is offered the quantification subelement.Quantize subelement, adjust the factor and add that the result after the coefficient side-play amount quantizes multiply by according to the quantization step that first kind of transfer pair answered.

Similar with embodiment one, in the decoding end adjustment factor of computational transformation A in advance, as conversion A during as more excellent mapping mode, decoding end is after carrying out inverse quantization, with the adjustment factor of inverse quantization result, and then carry out inverse transformation, thereby rebuild the coding side data divided by each self-corresponding conversion A.

If encoding code stream has comprised the adjustment factor of conversion A when coding, then decoding end includes the place of adjusting the factor from code stream, as sequence head, image head, slice header, macro block head, obtains the adjustment factor of conversion A.The occupation mode of the adjustment factor of conversion A is identical with said method, repeats no more here.

Embodiment three:

In the present embodiment, different to the compensation way of conversion A gained data with preceding two embodiment, employing be that the mode of adjusting point of quantification compensates.In addition, similar with embodiment one, the number range feature difference of two kinds of conversion in the present embodiment is that view data is respectively through two kinds of conversion and the corresponding difference that quantizes the back number range.Particularly, Fig. 6 for provide among the embodiment three to transform data processing method particular flow sheet.As shown in Figure 6, this method comprises:

Step 601～603, according to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion; At two kinds of conversion, the mean values scope behind the difference computational transformation; And calculate number range difference after two kinds of conversion and the corresponding quantification.

Operation in step 601～603 is identical with the operation of step 301～303 among the embodiment one, just repeats no more here.

Step 604 is determined the point of quantification side-play amount according to the difference in the step 603.

In the present embodiment, by adjusting conversion A corresponding quantitative point, conversion A gained data are compensated, thereby it is obtained and the corresponding to number range of conversion B, prepare for selecting more excellent mapping mode.

Conversion A corresponding quantitative point can be pre-set point of quantification, also can be the point of quantification that obtains from conversion B.

In video standard, adjacent two point of quantification corresponding quantization step be generally certain multiple relation, be specially: every k point of quantification, it is original 1/2nd that its corresponding quantitative step-length numerically becomes, and therefore can get multiple between the pairing quantization step of adjacent point of quantification and close and be

Multiple relation so be separated by between i pairing quantization step of point of quantification is

Again because quantization step with to utilize this quantization step to quantize the numerical value of back gained proportional, therefore, also can play the purpose that the value data scope is adjusted by the method for adjusting the point of quantification side-play amount.

According to the number range difference D ' behind the conversion A, the B that obtain in the step 603, can be compensated the needed point of quantification side-play amount of difference of this number range.Particularly, the concrete numerical value multiple of number range difference D ' correspondence is 2 ^{D '}, and after point of quantification added offset QP, the number that its corresponding quantitative step-length is former quantification step-length was

Doubly, make the two equate, then can obtain point of quantification offset QP=-|k * D ' |, wherein | | for rounding operator.Because the size of quantization step can influence the actual coding effect of data, the size that Δ QP can be on aforementioned calculation formula basis looks quantization step according to experimental data gained result is adjusted.

Certainly, the calculating of point of quantification side-play amount also can be adopted other formula, as long as can reflect the difference of the number range that obtains in the step 603.

Step 605, the data to after the encoding block C application conversion A acquisition conversion deduct definite quantification side-play amount with conversion A corresponding quantitative point, and the quantization step that finds according to adjusted point of quantification quantizes again.

By derivation Δ QP=-|k * D ' in the step 604 | process as seen, after point of quantification adjusted, the multiple pass of adjusted quantization step and former quantification step-length was 2 ^{D '},

Just adjusted quantization step becomes 2 ^{D '}QTAB1[s], wherein, QTAB1[s] be original quantization step.So, utilize the result after this revised quantization step quantizes to be

Number range with binary representation is Avr _B+ log ₂QTAB2[s].Obviously this number range is identical with the number range of gained data after conversion B and corresponding quantification, has realized making conversion A and the consistent purpose of the final number range of conversion B.

What so far, present embodiment provided finishes the transform data process flow.

Handle according to the encode transform data of side of said method, thereby select more excellent mapping mode among conversion A and the conversion B.If the final mapping mode of selecting is conversion A, the data that then send to decoding end are the data after conversion A and corresponding quantification and entropy coding etc. are handled.At this moment, need in advance according to step 601～604 computational transformation A corresponding quantitative point side-play amounts in decoding end, and this point of quantification side-play amount compensated, be specially and add Δ QP on the former corresponding quantitative point of the conversion A basis, utilize the point of quantification corresponding quantitative step-length after this change that data are carried out inverse quantization and inverse transformation again, to rebuild the coding side data exactly.

In above-mentioned decoding end step, when the point of quantification side-play amount is compensated, also can on the former corresponding quantitative point of conversion A basis, deduct Δ QP, utilize the point of quantification corresponding quantitative step-length after this change that data are carried out inverse quantization and inverse transformation again.Add or deduct Δ QP by quantization table characteristic decision on the former corresponding quantitative point of conversion A basis, be pre-set step, its purpose is that the coding side data can accurately recover in decoding end after conversion A handles.

Conversion A quantification side-play amount may adjust because of the difference of coded data in actual coding, so the adjustment factor of conversion A can be done suitable adjustment according to the characteristic of coded data when coding.Described coded data characteristic can be the actual average number range of data block after conversion A.And the quantification side-play amount of conversion A write in the encoding code stream.Be quantification side-play amount that conversion A corresponding quantitative point the is deducted number range of data after conversion A when can be depending on sequence of coding, piece image, a band or a macro block, the quantification side-play amount that is deducted writes encoding code stream.The quantification side-play amount of conversion A can write in the sequence head or image head or slice header or macro block head of code stream.The decoding end end then obtains the quantification side-play amount in sequence head from code stream or image head or slice header or the macro block head if know that the coding side code stream has comprised the point of quantification offset information, deducts the quantification side-play amount on the former corresponding quantitative point of conversion A basis.The former corresponding quantitative point of coding side conversion A corresponding quantitative point described here and decoding end conversion A can be pre-set point of quantification, also can be the point of quantification identical with conversion B.

Present embodiment also provides the another kind of embodiment of transform data processing unit, can be used to implement above-mentioned method shown in Figure 6.Fig. 7 is the concrete structure figure of this execution mode down conversion data processing equipment.As shown in Figure 7, this device comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit.Wherein, the conversion compensating unit comprises point of quantification correction subelement, varitron unit and quantizes subelement.

The first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer the number range difference unit.

Second value scope estimation unit is used for the required transformation matrix of second conversion according to default two kinds of conversion, the number range of estimated image data after through second conversion, and offer the number range difference unit.

The number range difference unit, be used for according to the number range and the corresponding separately point of quantification of first, second conversion of passing through respectively after first, second conversion, the estimated image data quantize the difference of back number range through first, second conversion and correspondence respectively, and offer the conversion compensating unit.

In the conversion compensating unit, point of quantification correction subelement, the difference that the number range that provides according to the number range difference unit is provided is determined the point of quantification side-play amount, and the point of quantification that first kind of transfer pair answered is deducted described point of quantification side-play amount, and adjusted result is offered the quantification subelement.The varitron unit is used for data to be transformed are used first kind of conversion, and the result after the conversion is offered the quantification subelement.Quantize subelement, the adjusted point of quantification that provides according to the described point of quantification correction subelement result after to conversion is provided quantizes.

In the specific implementation process of above-mentioned present embodiment, the difference of the number range behind conversion A, B is according to the mean values range computation.In fact, the calculating of this difference can adopt the mode of embodiment two to carry out, promptly at each pixel in the sub-piece behind conversion A, and the difference of difference evaluation scope.Correspondingly, when in step 604, adjusting conversion A corresponding quantitative point, also adjust at each pixel in the sub-piece behind conversion A, particularly,

Then, in step 605,, quantize according to adjusted point of quantification corresponding quantitative step-length according to different pixels.Like this, can make the result of compensation more accurate, make the consistent degree of the number range of gained data after final two kinds of conversion and the corresponding quantification higher, thereby can select more excellent conversion more effectively, further improve code efficiency.

Correspondingly, when coding side after to select conversion A be more excellent mapping mode, each pixel data corresponding quantitative point side-play amount after decoding end obtains conversion A according to said method in advance, and the data that will treat inverse quantization separately the corresponding quantitative point add side-play amount, utilize the point of quantification corresponding quantitative step-length after changing to carry out inverse quantization and inverse transformation again, to rebuild the coding side data exactly.

When quantizing side-play amount and need write code stream, can copy preceding method to carry out the writing of code stream, parsing and point of quantification side-play amount in coding side and decoding end processing method to the former corresponding quantitative point of conversion A.

In above-mentioned three embodiment, data after having adopted two kinds of different modes to conversion respectively compensate, thereby reach the purpose of adjusting its number range, particularly, a kind of mode is the data after the conversion to be multiply by one adjust the factor, and another kind of mode is to change the point of quantification that this transfer pair is answered.In fact, above-mentioned dual mode can be combined the compensation of carrying out data after the conversion.For example, the data after the conversion be multiply by one adjust factor f, simultaneously point of quantification is offset s, purpose is to allow following relation set up: f * Avr _A+ s=Avr _B, (1＜=i＜=n, 1＜=j＜=n).Because Avr _AAnd Avr _BBe priori data, f and s can be preset value, and value is not unique, and concrete f and s can look Avr _AAnd Avr _BSize of data relation and encoding and decoding realization condition and set.If f and s need adjust according to image coded data, then f and s can write the first-class place of sequence head, image head, slice header, macro block in the encoding code stream.If encoding code stream contains f and s, then decoding end is from the relevant position of encoding code stream, and for example sequence head, image head, slice header, macro block are first-class, resolve and obtain f and s.

Certainly, also all pixel datas after the sub-piece conversion can be multiply by the adjustment factor of a correspondence respectively, simultaneously point of quantification is offset s _{(i, j)}, purpose is to allow following relation set up: f _{(i, j)}* Aij+s _{(i, j)}=Avr _B, (1＜=i＜=n, 1＜=j＜=n).Wherein, f _{(i, j)}And s _{(i, j)}Can look Avr _AAnd Avr _BSize of data relation and encoding and decoding realization condition and set.If f _{(i, j)}And s _{(i, j)}Need adjust according to image coded data, then f _{(i, j)}And s _{(i, j)}Can write the first-class place of sequence head, image head, slice header, macro block in the encoding code stream.

If to compensating, can set in advance identical f in decoding end through conversion A gained data through aforesaid way _{(i, j)}And s _{(i, j)}After the coded data that receives correspondent transform A, before these data are carried out inverse quantization, adjust conversion A corresponding quantitative point, on the basis of original point of quantification, add s _{(i, j)}, treat the adjustment factor f of the data of inverse quantization simultaneously divided by correspondence _{(i, j)}, utilize the point of quantification corresponding quantitative after changing to compensate again to carrying out inverse quantization, with accurate reconstruction coding side data divided by the inverse quantization data for the treatment of after the adjustment factor.

If encoding code stream contains f _{(i, j)}And s _{(i, j)}, then decoding end is from the relevant position of encoding code stream, and for example sequence head, image head, slice header, macro block are first-class, resolve and obtain f _{(i, j)}And s _{(i, j)}

To method that should above-mentioned two kinds of compensation way combinations, the present invention also provides corresponding apparatus structure.This apparatus structure and similar shown in Figure 7, difference are that conversion compensating unit wherein comprises to be adjusted the factor and point of quantification correction subelement, varitron unit, conversion reprocessing subelement and quantize subelement.All the device with shown in Figure 7 is identical for the 26S Proteasome Structure and Function of other first number range estimation unit, second value scope estimation unit, number range difference unit.

In this conversion compensating unit, adjust the factor and point of quantification correction subelement, the difference that the number range that provides according to the number range difference unit is provided determines to adjust the factor and point of quantification side-play amount, and the point of quantification that first kind of transfer pair answered deducted described point of quantification side-play amount, adjusted point of quantification is offered described quantification subelement, the adjustment factor of determining is offered conversion reprocessing subelement.The varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered conversion reprocessing subelement.Conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment factor, and the result is offered the quantification subelement.Quantize subelement, quantize multiply by the result who adjusts after the factor according to described adjusted point of quantification corresponding quantitative step-length.

In above-mentioned three embodiment, conversion A can adopt different quantization tables with conversion B, and just the quantization step that finds according to point of quantification is different, but for the identical situation of quantization table, adopts aforesaid way bigger for the loss of data precision.

At the identical situation of quantization table, the present invention proposes the execution mode of embodiment four, for a kind of conversion wherein, redesign one cover quantization table, thus make after two kinds of conversion and the corresponding quantification number range of data consistent.

Embodiment four:

Different with aforementioned three embodiment, in the present embodiment, the number range feature difference of two kinds of conversion is that view data is respectively through the difference of number range after two kinds of conversion.

Fig. 8 is the particular flow sheet of transform data processing method in the embodiment of the invention four.As shown in Figure 8, this method comprises:

Step 801～802, according to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion; At two kinds of conversion, the mean values scope behind the difference computational transformation;

Operation in step 801～802 is identical with the operation of step 301～302 among the embodiment one, and difference is, carries out aforesaid operations respectively at different point of quantification.That is to say,, calculate the mean values scope of this macro block after two kinds of conversion: Avr respectively for all different macro blocks _A(n) and Avr _B(n), wherein, n is the point of quantification index.

Step 803 is calculated the number range difference after two kinds of conversion.

At the number range difference after different two kinds of conversion of point of quantification calculating, particularly, Δ d (n)=Avr _B(n)-Avr _A(n).

In fact, as previously mentioned, because in the coding and decoding video theory, use the uniform quantization method usually, just every several point of quantification, quantization step be reduced to original half, so the numerical value of Δ d (n) also has certain regularity.Based on this, can do a simplification here.Particularly, establish every m point of quantification, quantization step be reduced to original half, then can select to calculate continuous m Δ d (n), for example, this step is only calculated Δ d (1), Δ d (2), L, this m difference that Δ d (m) has been an assumed calculation below.

Step 804 is provided with each point of quantification corresponding quantitative step-length and displacement figure place, and forms quantization table.

In this step, when quantization step is set, carry out according to the difference that calculates in the step 803.Particularly, corresponding to the m that calculates in the step 803 difference, the 1st to m the pairing quantization step of point of quantification is set is: 2 ^{Δ d (1)}* R ... ... .2 ^{Δ d (m)}* R, wherein R is a constant coefficients value, and purpose is to allow the value of quantization step be an integer and improve computational accuracy (molecule is big, and nature can keep more information), but considers the restriction of hardware store scope, and this R value has its upper limit.According to the uniform quantization principle, just can release the quantization step of other point of quantification by the 1st to m the pairing quantization step of point of quantification.

Certainly, when quantization step is set, also can all quantization steps be provided with, concrete n point of quantification corresponding quantitative step-length is 2 ^{Δ d (n)}* R.

Being set when the displacement figure place is set is a fixed value t.When this t was big, the coding effect was relatively poor, the rate of compression coding height; When this t hour, the coding effect is better, rate of compression coding is low.Therefore, the concrete value of t need be taken all factors into consideration coding effect and rate of compression coding, finds the corresponding t of a better balance point.

Displacement figure place and quantization step according to above-mentioned setting form quantization table, and this process is identical with existing implementation, just repeats no more here.

Step 805 receives data to be transformed, and uses the data after conversion A obtains conversion, quantizes according to the data of above-mentioned quantization table after to conversion.

After above-mentioned processing, the number range of coding side data data after conversion A and above-mentioned quantization table quantize can reach basically identical with the number range after conversion B and corresponding quantification.

Why present embodiment realizes that the principle of goal of the invention is identical with embodiment one, particularly, because in the quantizing process, quantization step is directly proportional with quantification back data, therefore the process that conversion A gained data is quantized according to above-mentioned quantization step with that conversion A gained data be multiply by a process of adjusting the factor is of equal value substantially among the embodiment one, all is to multiply by 2 in fact ^{Δ d (n)}Relation.Therefore, present embodiment also can realize adjusting number range so that its with conversion B and corresponding quantification after the consistent purpose of number range.

Correspondingly, also to design and the corresponding inverse quantization table of the quantization table of coding side in decoding end.In the inverse quantization table, consider in the method for coding side according to uniform quantization, the quantization step of n point of quantification is the twice of the quantization step of (n+m) individual point of quantification, the utilization displacement compensates the multiple relation between quantization step, just displacement figure place+1 of displacement figure place=(n+m) individual point of quantification of n point of quantification when inverse quantization in the inverse quantization table.The displacement figure place of the 1st to m inverse quantization point can be made as k, and the displacement figure place of m+1 to 2m inverse quantization point just should be k-1 mutually so.As for, the setting of k value, also be to consider the memory range of hardware and decide, principle is identical with the situation of aforementioned consideration hardware store scope.

Below, according to the setting of above-mentioned displacement figure place, the inverse quantization step-length of calculating the inverse quantization table is IQ (n), wherein n is a point of quantification.IQ (n) needs to satisfy relation Wherein, What reflect is the variation of displacement figure place, and promptly every m point of quantification, the displacement figure place increases by 1.According to this formula as can be known

Wherein h is a constant constant, and its value is the bigger the better, but will be subjected to the restriction of hardware store scope equally.

Set up the inverse quantization table in the manner described above,, use the data of conversion A after carrying out inverse transformation, to use the inverse quantization table of this foundation just can the restoration and reconstruction data like this in decoding end.

Present embodiment also provides a kind of embodiment of transform data processing unit, can be used to implement above-mentioned method shown in Figure 8.Fig. 9 is the concrete structure figure of this execution mode down conversion data processing equipment.As shown in Figure 9, this device comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit.Wherein, the conversion compensating unit comprises that quantization table sets up subelement, varitron unit and quantize subelement.

The first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer the number range difference unit.

Second value scope estimation unit is used for the required transformation matrix of second conversion according to default two kinds of conversion, the number range of estimated image data after through second conversion, and offer the number range difference unit.

Described number range difference unit is used for estimating the difference through number range after two kinds of conversion according to the number range of passing through respectively after first, second conversion, and this difference is offered described conversion compensating unit as the number range feature difference.

In the conversion compensating unit, quantization table is set up subelement, and being used for according to described number range feature difference is that the corresponding quantitative table is set up in first kind of conversion, and offers the quantification subelement.The varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered the quantification subelement.Quantize subelement, the quantization table of answering according to described first kind of transfer pair is the result after to first kind of conversion quantize.

In the embodiment of the invention described above, all, the data through conversion A are example so that being compensated, the specific embodiment of the present invention is described, in fact, also can the data through conversion B be compensated, its execution mode is identical, just conversion A and the pairing parameter of conversion B is exchanged to get final product in corresponding formula, just repeats no more here.

The method that above-mentioned transform data is handled can be used for coding method, and Figure 10 is the coding method schematic flow sheet that the embodiment of the invention five provides, and this coding method may further comprise the steps:

A1, reception data to be transformed;

Described data to be transformed comprise the image block residual error that obtains through prediction.

A2, described data to be transformed are carried out first conversion, obtain the data after first conversion; Described data to be transformed are carried out second conversion, obtain the data after second conversion;

Described first conversion here is the conversion of 4x4 size;

Described second conversion here is the conversion of 8x8 size;

A3, determine the adjustment parameter of first conversion according to the data after the data after first conversion and second conversion, according to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion.

Data after first conversion here and the data after second conversion comprise number range after aforesaid first conversion and the number range after second conversion, for convenience of description, adopt the data after first conversion and the call of the data after second conversion.

Adjust parameter and can comprise that the data after described first conversion of then described parameter adjustment according to described adjustment parameter and described second conversion comprise: will multiply by the described adjustment factor through the data after first conversion according to the data after described first conversion and the definite factor of adjusting of the data after second conversion; Quantize multiply by the data of adjusting after the factor according to the corresponding quantitative step-length.

Adjust parameter and can comprise the quantification side-play amount of determining according to the data after the data after first conversion and second conversion, the parameter of described second conversion is the point of quantification of second conversion, and the data after described first conversion of described parameter adjustment according to the described adjustment parameter and second conversion comprise: the point of quantification that second transfer pair is answered deducts the point of quantification of described quantification side-play amount as first conversion; Utilize the point of quantification of adjusted first conversion that the data after first conversion are quantized.

Adjusting parameter can comprise the quantification side-play amount of determining according to the data after the data after first conversion and second conversion and adjust the factor; The parameter of described second conversion is the point of quantification that second transfer pair is answered, data after described first conversion of then described parameter adjustment according to described adjustment parameter and described second conversion comprise: will multiply by the described adjustment factor through the data after first conversion, and the point of quantification that second transfer pair is answered is deducted the point of quantification of described quantification side-play amount as first conversion; Utilize described first conversion point of quantification quantize multiply by the data of adjusting after the factor.

Adjusting parameter can also comprise the adjustment coefficient of determining according to the data after the data after first conversion and second conversion and quantize the displacement side-play amount; The parameter of described second conversion comprises, to the carry digit of the point of quantification of second transform data; Data after described first conversion of then described parameter adjustment according to the described adjustment parameter and second conversion comprise: will multiply by described adjustment coefficient through the data after first conversion, and the displacement figure place of utilizing the point of quantification that first transfer pair answers quantizes with the data of adjusting after the multiplication to described; The displacement figure place of the point of quantification that described first transfer pair is answered is the displacement figure place and the described quantification displacement side-play amount sum of second transfer pair point of quantification of answering.

Described adjustment parameter can comprise the adjustment factor and the coefficient side-play amount of determining according to the data after the data after first conversion and second conversion, data after described first conversion of then described parameter adjustment according to the described adjustment parameter and second conversion comprise: will multiply by the described adjustment factor through the data after first conversion, and add described coefficient side-play amount and quantize.

A4, described adjustment parameter is write in the encoding code stream.

Described adjustment parameter is write in the sequence head or image head or slice header or macro block head in the encoding code stream, use for decoding end.

After the number range of data after first conversion after the above-mentioned processing and second conversion number range of data roughly just as, influence that can more effective reflection transfer pair data when coding is selected the more excellent conversion of effect, and then is improved code efficiency.

Figure 11 is the coding/decoding method schematic flow sheet that the embodiment of the invention six provides, and this method comprises:

B1, receive code stream, described code stream is decoded, obtain the data after first conversion and adjust parameter;

Data after first conversion are meant if current decoded image blocks is judged as uses first conversion, then the data of entropy decoding back gained image block residual error after the first conversion inverse transformation;

Described first conversion here is the inverse transformation of 4x4 size;

Described second conversion here is the inverse transformation of 8x8 size;

Described adjustment parameter is corresponding with the adjustment parameter that writes code stream when encoding.

B2, according to the data after described first conversion of the parameter adjustment of described adjustment parameter and described second conversion.

This step comprises several different methods, is respectively:

(1) determine the adjustment factor to the data after first conversion according to the parameter of described adjustment parameter and described second conversion: the data of the quantization step of answering according to first transfer pair after to first conversion are carried out inverse quantization, and the merchant of the inverse quantization result and the described adjustment factor is carried out inverse transformation.

(2) if the adjustment parameter that receives is for quantizing side-play amount, the parameter of then described second conversion is the point of quantification that second transfer pair is answered, the point of quantification of first conversion is that the point of quantification that described second transfer pair is answered deducts described quantification side-play amount, the data adjustment of described point of quantification of answering according to described quantification side-play amount and second transfer pair after to first conversion comprises: the data after utilizing point of quantification that described first transfer pair answers to described first conversion are carried out inverse quantization, and the inverse quantization result is carried out inverse transformation.

(3) comprise and quantize side-play amount and if adjust parameter the adjustment factor of the data after first conversion; The parameter of then described second conversion is the point of quantification that second transfer pair is answered, and the point of quantification of first conversion is that the point of quantification of described second conversion deducts described quantification side-play amount; Data after described first conversion of described parameter adjustment according to described adjustment parameter and described second conversion comprise: the data after utilizing point of quantification that described first transfer pair answers to first conversion are carried out inverse quantization; Merchant to the inverse quantization result and the described adjustment factor carries out inverse transformation.

(4) if described adjustment parameter comprises the adjustment coefficient of first conversion and to the quantification of first transform data displacement side-play amount; The parameter of described second conversion comprises, to the carry digit of the point of quantification of second transform data; Data after described first conversion of then described parameter adjustment according to described adjustment parameter and described second conversion comprise: the displacement figure place of the point of quantification that described second transfer pair is answered deducts the displacement figure place of the point of quantification that described quantification displacement side-play amount answers as first transfer pair; The displacement figure place of utilizing the point of quantification that described first transfer pair answers is carried out inverse quantization to the data that receive after first conversion; Merchant to inverse quantization result and described adjustment coefficient carries out inverse transformation.

(5) according to the definite adjustment factor of the parameter of described adjustment parameter and described second conversion to first transform data; According to the definite coefficient side-play amount of the parameter of described adjustment parameter and described second conversion to first transform data; Data after described first conversion of described parameter adjustment according to described adjustment parameter and described second conversion comprise: the quantization step of answering according to described first transfer pair carries out inverse quantization to the data that receive, and the inverse quantization result deducted described coefficient side-play amount, the merchant of the described result of subtracting each other and the described adjustment factor is carried out inverse transformation.

Corresponding with coding side, described adjustment parameter from code stream sequence head or image head or slice header or macro block head in obtain.

Figure 12 is the code device structural representation that the embodiment of the invention seven provides, and this code device comprises:

The Data Receiving unit is used to receive data to be transformed;

Converter unit is used for the data described to be transformed that receiving element receives are carried out first conversion, obtains the data after first conversion; Described data to be transformed are carried out second conversion, obtain the data after second conversion;

First adjustment unit is determined to adjust parameter according to the parameter of second conversion, data after first conversion and the data after second conversion, according to the data after described first conversion of the parameter adjustment of the described adjustment parameter and second conversion;

R/w cell is used for described adjustment parameter is write encoding code stream.

Figure 13 is the decoding device structural representation that the embodiment of the invention eight provides, and this decoding device comprises:

The code stream receiving element is used to receive code stream;

Decoding unit is decoded to described code stream, obtains the data after first conversion and adjusts parameter;

Second adjustment unit is used for according to the data after parameter adjustment first conversion of the described adjustment parameter and second conversion.

Above-mentioned encoding apparatus and decoding apparatus can be formed cover system use, are not giving unnecessary details here.

Embodiment nine: a kind of coding method

Embodiment nine provides a kind of coding method, can ask for adaptively at coding side to quantize side-play amount (aforesaid adjustment parameter), and as shown in figure 14, described method comprises:

C1, receive data to be transformed and coding parameter information;

C2, described data to be transformed are carried out first conversion, obtain the data after first conversion;

C3, determine to adjust parameter according to the parameter of the coding parameter information and second conversion, according to the data after described first conversion of the parameter adjustment of described adjustment parameter and described second conversion;

C4, described adjustment parameter is write in the encoding code stream.

Introduce in detail below.

If the yardstick of conversion A is 4x4, the yardstick of conversion B is 8x8.

At first receive data to be transformed and coding parameter information in the present embodiment.Described data to be transformed are carried out first conversion, obtain the data after first conversion, according to the coding parameter information and second transformation parameter determine to adjust parameter.Second transformation parameter described here is the pairing point of quantification of second conversion, and described adjustment parameter is for quantizing side-play amount.

The described information of coding parameter comprises the infra-frame prediction image block number that uses first conversion, use the infra-frame prediction image block number of second conversion, all use the number of infra-frame prediction image block, use the infra-frame prediction image block number of first conversion in P frame or the B frame, use the infra-frame prediction image block number of first conversion in P frame and the B frame, use the number of predicted picture piece in second transform frame in P frame or the B frame, use the number of predicted picture piece in second transform frame in P frame and the B frame, use the number of infra-frame prediction image block in P frame and the B frame, use the number of the image block of skip mode or Direct Model in the P frame, use the number of the image block of skip mode and Direct Model in the P frame, use the number of the image block of skip mode or Direct Model in the B frame, use the number of the image block of skip mode and Direct Model in the B frame, use the number of inter-frame forecast mode image block in the P frame, or use in the number of inter-frame forecast mode image block one or more in the B frame.Wherein skip mode is a common technology in the coding and decoding video, its implication for according to encoded or the decoded picture block message obtain the present image block motion vector, and do not have coded residual in this image block; The Direct Model implication for according to encoded or the decoded picture block message obtain the present image block motion vector, but contain coded residual in this image block.

Remember that the described infra-frame prediction image block number of first conversion that uses is img-＞intra4x4num, the described number of infra-frame prediction image block that uses is img-＞intra_num, use in the described P frame skip mode and or the number of the image block of Direct Model be img-＞pskip_num, use the number of infra-frame prediction image block to be img-＞p_intra in the described P frame, use the number of inter-frame forecast mode image block to be img-＞pnum in the described P frame, the pairing point of quantification of described second conversion is img-＞qp, described quantification side-play amount img-＞qp_shift.

Be used for determining to quantize side-play amount three of coding side storages in the present embodiment, every table comprises 64 elements, these three souvenirs are QPshift_table[0], QPshift_table[1] and QPshift_table[2] three arrays, comprise 64 elements in each array, the form of expression is as follows:

QPshift_table[0]＝{0，1，1，1，1，2，2，2，2，3，3，3，3，4，4，4，4，5，5，5，5，6，6，6，6，7，8，8，8，8，8，8，8，8，9，9，9，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10，10}

QPshift_table[1]＝{0，1，1，1，1，2，2，2，2，3，3，3，3，4，4，4，4，5，5，5，5，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6，6}

QPshift_table[2]＝{0，1，1，1，2，2，3，3，3，4，4，4，5，5，6，6，7，8，8，9，9，10，10，10，10，10，11，11，11，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，12，13，13，13，13，13，13，13，13，13，13，13，13}

When using the adaptive block transformation technology, if current encoded image is the I frame and is first width of cloth image of current sequence or present image group, note img-＞intra4x4num, img-＞intra_num, img-＞pskip_num, img-＞p_intra and img-＞pnum value are 0, use QPshift_table[0] determine img-＞QP_shift, the method for determining is: img-＞QP_shift=QPshift_table[0] [img-＞qp].Here said image sets is the combination that has comprised in some I frames, P frame, B frame or the above-mentioned several types.

If current encoded image is an I frame and for first width of cloth image of current sequence or present image group but also do not have in current sequence or the present image group and then set two parameters with P frame image encoded, be designated as the first determined value v1 and the second determined value v2, the computational methods of v1 and v2 are as follows:

v1＝img-＞intra4x4num*5.0/img-＞intra_num/(img-＞QP_shift+12)

v2＝0.2

Symbol wherein ' * ' represent multiplication, symbol '/' represent division.

Obtain obtaining quantizing side-play amount img-＞QP_shift according to following logic determines after v1 and the v2:

(1) if v1 less than 0.3 and v2 less than 0.4, perhaps v1 less than 0.4 and v2 less than 0.15, perhaps v1 less than 0.5 and v2 less than 0.1, then the value of img-＞QP_shift is an index value with img-＞qp, the table QPshift_table[1] in search acquisition;

If not satisfying, condition (1) then do not carry out to judge:

(2) if v1 greater than 0.5 or v2 greater than 0.5 or v1*v2 greater than 0.2, then the value of img-＞QP_shift is an index value with img-＞qp, the table QPshift_table[2] in search acquisition;

The value of img-＞QP_shift is an index value with img-＞qp if condition (1), (2) do not satisfy then, at table QPshift_table[0] in search acquisition.

If current encoded image is not then set two parameters for existing in first width of cloth image of current sequence or present image group and current sequence or the present image group with P frame image encoded, be designated as the first determined value v1 and the second determined value v2, the computational methods of v1 and v2 are as follows:

v1＝img-＞intra4x4num*5.0/img-＞intra_num/(img-＞QP_shift+12)；

v2＝img-＞pskip_num*1.0/img-＞pnum

Symbol wherein ' * ' represent multiplication, symbol '/' represent division.

Obtain obtaining quantizing side-play amount img-＞QP_shift according to following logic determines after v1 and the v2:

(1) if v1 less than 0.3 and v2 less than 0.4, perhaps v1 less than 0.4 and v2 less than 0.15, perhaps v1 less than 0.5 and v2 less than 0.1, then the value of img-＞QP_shift is an index value with img-＞qp, the table QPshift_table[1] in search acquisition;

If not satisfying, condition (1) then do not carry out to judge:

(2) if v1 greater than 0.5 or v2 greater than 0.5 or v1*v2 greater than 0.2, then the value of img-＞QP_shift is an index value with img-＞qp, the table QPshift_table[2] in search acquisition;

The value of img-＞QP_shift is an index value with img-＞qp if condition (1), (2) do not satisfy then, at table QPshift_table[0] in search acquisition.

Obtain quantizing after side-play amount img-＞QP_shift, the form of storing in advance according to coding side obtains the pairing weight coefficient lambda of first conversion in the rate-distortion optimization model (rate-distortion optimization) again.This form is designated as QP_lambda_table_4x4, includes 16 elements in the present embodiment.Use the lambda of in-frame encoding picture piece to be designated as img-＞lambda4x4I in first kind of conversion, use the lambda of the image block of inter-frame forecast mode to be designated as img-＞lambda4x4p, the computational methods of img-＞lambda4x4I and img-＞lambda4x4p are:

img-＞lambda4x4I＝QP_lambda_table_4x4[img-＞QP_shift]

img-＞lambda4x4p＝img-＞lambda4x4I*0.9

The form of expression of QP_lambda_table_4x4 is as follows:

QP_lambda_table_4x4[16]＝

{1.0，1.1，1.3，1.5，1.8，2.1，2.6，3.1，3.5，4.0，4.6，5.1，5.7，6.3，100，100}

If current image to be encoded is the P frame, then img-＞QP_shift is made as 5, and img-＞lambda4x4p is made as 1.6.

If current image to be encoded is the B frame, then img-＞QP_shift is made as 5.

After obtaining img-＞QP_shift by said method, calculate the point of quantification of first conversion according to the parameter (i.e. the point of quantification of second conversion) of second conversion, concrete grammar deducts the point of quantification of described quantification side-play amount as first conversion for the point of quantification with second conversion.Utilize the point of quantification of first conversion of calculating gained that the data after first conversion are quantized.To quantize side-play amount img-＞QP_shift writes in the encoding code stream.

Present embodiment ten is also corresponding to provide a kind of code device, can be used to carry out the coding method that embodiment nine provides, and as shown in figure 15, described code device comprises:

The 3rd receiving element is used to receive data to be transformed and coding parameter information;

The 3rd converter unit is used for the data to be transformed that described the 3rd receiving element receives are carried out first conversion, obtains the data after first conversion;

The 3rd adjustment unit, be used for the information of coding parameter that receives according to the 3rd receiving element and the parameter of second conversion and determine to adjust parameter, the data after first conversion that obtains according to described the 3rd converter unit of the parameter adjustment of described adjustment parameter and described second conversion;

The 3rd r/w cell is used for the adjustment parameter that described the 3rd adjustment unit obtains is write encoding code stream.

The 3rd receiving element receives data to be transformed and coding parameter information in the present embodiment.The 3rd converter unit carries out first conversion to described data to be transformed, obtains the data after first conversion, and the 3rd adjustment unit is according to the coding parameter information and second transformation parameter determine to adjust parameter.Second transformation parameter described here is the pairing point of quantification of second conversion, and described adjustment parameter is for quantizing side-play amount.

The described information of coding parameter comprises the infra-frame prediction image block number that uses first conversion, use the infra-frame prediction image block number of second conversion, all use the number of infra-frame prediction image block, use the infra-frame prediction image block number of first conversion in P frame or the B frame, use the infra-frame prediction image block number of first conversion in P frame and the B frame, use the number of predicted picture piece in second transform frame in P frame or the B frame, use the number of predicted picture piece in second transform frame in P frame and the B frame, use the number of infra-frame prediction image block in P frame and the B frame, use the number of the image block of skip mode or Direct Model in the P frame, use the number of the image block of skip mode and Direct Model in the P frame, use the number of the image block of skip mode or Direct Model in the B frame, use the number of the image block of skip mode and Direct Model in the B frame, use the number of inter-frame forecast mode image block in the P frame, or use in the number of inter-frame forecast mode image block one or more in the B frame.Wherein skip mode is a common technology in the coding and decoding video, its implication for according to encoded or the decoded picture block message obtain the present image block motion vector, and do not have coded residual in this image block; The Direct Model implication for according to encoded or the decoded picture block message obtain the present image block motion vector, but contain coded residual in this image block.

The 3rd adjustment unit calculates the quantification side-play amount according to the point of quantification of the above-mentioned coded message and second conversion, calculate the point of quantification of first conversion again according to the point of quantification that quantizes the side-play amount and second conversion, and the data of the point of quantification of utilizing first conversion that obtains after to first conversion quantize (data after first conversion that obtains according to described the 3rd converter unit of the parameter adjustment of described adjustment parameter and described second conversion).The content that the concrete step of calculating can reference example nine be described.

The 3rd r/w cell will quantize side-play amount and write in the encoding code stream, carry out decode operation so that decoding time obtains to quantize side-play amount.

Embodiment nine and embodiment ten obtained adjustment parameter (quantification side-play amount) according to the coding parameter information and second transformation parameter, makes coding side have better flexibility.Because coding side will calculate the adjustment parameter of gained and write in the encoding code stream, so decoding end only needs to obtain this adjustment parameter from code stream.The described method of this embodiment has been taken into account coding efficiency can additionally not increase the decoding end burden simultaneously yet.

Being preferred embodiment of the present invention only below, is not to be used to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any modification of being done, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (28)

1. a transform data processing method is characterized in that, this method comprises:

According to default two kinds of transformation matrixs that conversion is required, the estimated image data are respectively through the number range after two kinds of conversion;

According to the number range after described two kinds of conversion, estimate the number range feature difference of two kinds of conversion, the number range feature difference of described two kinds of conversion be view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, or view data is respectively through the difference of number range after two kinds of conversion;

Receive data to be transformed, and these data are used first kind of conversion in described two kinds of conversion, and according to the number range feature difference of estimating compensate after this first kind of conversion data so that the number range behind the number range behind first kind of transform and quantization and the second kind of transform and quantization be consistent.

2. method according to claim 1, it is characterized in that, further the two kinds of conversion quantization step separately that finds according to view data corresponding quantitative point is estimated the number range feature difference of described two kinds of conversion, wherein, the number range feature difference of described two kinds of conversion is that view data is respectively through two kinds of conversion and the corresponding difference that quantizes the back number range.

3. method according to claim 2 is characterized in that, the data that described number range feature difference according to estimation compensates after this first kind of conversion are:

According to estimating the definite factor of adjusting of number range feature difference, will multiply by this adjustment factor through the data after first kind of conversion;

Quantize multiply by the data of adjusting after the factor according to the corresponding quantitative step-length.

4. method according to claim 3 is characterized in that, described number range feature difference according to estimation is determined to adjust factor q and is:

Wherein, D ' is through two kinds of conversion and the corresponding difference that quantizes the back number range.

5. method according to claim 3, it is characterized in that, this method further comprises: decoding end determines to adjust the factor according to the number range feature difference in advance, when the data of decoding after first kind of conversion process, the quantization step of answering according to first kind of transfer pair carries out inverse quantization to the data that receive, and the merchant of the inverse quantization result and the described adjustment factor is carried out inverse transformation.

6. method according to claim 2 is characterized in that, the data that described number range feature difference according to estimation is handled after the conversion are:

Determine the point of quantification side-play amount according to the number range feature difference of estimating, and the point of quantification that this transfer pair is answered is deducted described point of quantification side-play amount;

Data after utilizing adjusted point of quantification to first kind of conversion quantize.

7. method according to claim 6, it is characterized in that, described difference according to the number range of estimating determines that point of quantification offset QP is Δ QP=-|k * D ' |, wherein, D ' is through two kinds of conversion and the corresponding difference that quantizes the back number range, k is that quantization step reduces a half corresponding quantitative point side-play amount, | | for rounding operator.

8. method according to claim 6, it is characterized in that, this method further comprises: decoding end is determined the point of quantification side-play amount according to the number range feature difference in advance, when the data of decoding after first kind of conversion process, the corresponding quantitative point is added the above point of quantification side-play amount, and utilize this adjusted point of quantification that the data that receive are carried out inverse quantization, and the inverse quantization result is carried out inverse transformation.

9. method according to claim 2 is characterized in that, the data that described number range feature difference according to estimation compensates after this first kind of conversion are:

Determine to adjust the factor and point of quantification side-play amount according to the number range feature difference of estimating;

To multiply by definite adjustment factor through the data after first kind of conversion, and the point of quantification that first kind of transfer pair answered will be reduced described point of quantification side-play amount;

Utilize adjusted point of quantification to quantize to multiply by the data of adjusting after the factor.

10. method according to claim 9, it is characterized in that, this method further comprises: decoding end determines to adjust the factor and point of quantification side-play amount according to the number range feature difference in advance, when the data of decoding after first kind of conversion process, the corresponding quantitative point is added the above point of quantification side-play amount, and utilize this adjusted point of quantification that the data that receive are carried out inverse quantization, and the merchant of the inverse quantization result and the described adjustment factor is carried out inverse transformation.

11. method according to claim 2 is characterized in that, the data that described number range feature difference according to estimation compensates after this first kind of conversion are:

Displacement side-play amount when determining to adjust coefficient and quantification according to the number range feature difference of estimating;

To multiply by this adjustment coefficient through the data after first kind of conversion, and the figure place that will be shifted increases described displacement side-play amount when quantizing;

Wherein, described adjustment coefficient is q '=2 ^{D '}* 2 ⁿ, the displacement side-play amount is n, and wherein, n is a positive integer, and D ' is through two kinds of conversion and the corresponding difference that quantizes the back number range.

12. method according to claim 11, it is characterized in that, this method further comprises: the displacement side-play amount when decoding end is determined to adjust coefficient and quantification according to the number range feature difference in advance, when the data of decoding after first kind of conversion process, the displacement figure place of corresponding point of quantification is deducted described displacement side-play amount, and utilize this adjusted point of quantification that the data that receive are carried out inverse quantization, and the merchant of inverse quantization result and described adjustment coefficient is carried out inverse transformation.

13. method according to claim 2 is characterized in that, the data that described number range feature difference according to estimation compensates after this first kind of conversion are:

Determine to adjust the factor and coefficient side-play amount according to the number range feature difference of estimating;

To multiply by definite adjustment factor through the data after first kind of conversion, add described coefficient side-play amount;

The quantization step that utilizes first transfer pair to answer is adjusted the factor and is added that the data after the coefficient side-play amount quantize multiply by.

14., it is characterized in that described estimated image data through two kinds of conversion and the corresponding difference that quantizes the back number range are respectively according to arbitrary described method in the claim 2 to 13:

For two kinds of conversion, the mean values scope behind the difference computational transformation

With

Wherein, to be view data divide in the sub-piece that obtains the number range of the capable j row of i pixel data after first kind of conversion according to the yardstick of first kind of conversion to Aij, to be view data divide in the sub-piece that obtains the number range of the capable j row of i pixel data after second kind of conversion according to the yardstick of second kind of conversion in two kinds of conversion to Bij, the yardstick of first kind of conversion is n * n, and the yardstick of second conversion is m * m;

According to mean values scope after two kinds of conversion and corresponding quantitative step-length separately, the computed image data are respectively through two kinds of conversion and the corresponding difference D '=(Avr that quantizes the back number range _B+ log ₂QTAB2[n])-(Avr _A+ log ₂QTAB1[n]), wherein, QTAB1[n] and QTAB2[n] be respectively the quantization step that the point of quantification of answering according to first, second kind transfer pair finds.

15., it is characterized in that described estimation video data through two kinds of conversion and the corresponding difference that quantizes the back number range is respectively according to arbitrary described method in the claim 2 to 13: calculate in described two kinds of conversion the mean values scope after second kind of conversion

According to the mean values scope after the second conversion kind and the number range of dividing sub-piece all pixel datas after first kind of conversion that obtain according to the yardstick of conversion, and two kinds of conversion corresponding quantitative point separately, calculate described sub-piece through first kind of conversion and the corresponding number range of the capable j row of back i pixel data and the difference d ' of quantizing through the mean values scope after second kind of conversion and the corresponding quantification _{(i, j)}=(Avr _B+ log ₂QTAB2[n])-(Aij+log ₂QTAB1[n]), wherein, Aij is the number range of the described sub-piece capable j row of i pixel data after first kind of conversion, QTAB1[n] and QTAB2[n] be respectively the quantization step that the point of quantification of answering according to first, second transfer pair finds.

16. method according to claim 1 is characterized in that,

The operation of estimating described number range is carried out respectively at k point of quantification at least continuously, and wherein every k point of quantification, quantization step is reduced to original half;

The number range feature difference of described two kinds of conversion is: at described k point of quantification at least continuously, through the difference DELTA d (s) of number range after two kinds of conversion, described s is the point of quantification index to view data respectively;

After estimating the number range feature difference of described two kinds of conversion, receive data to be transformed before, the displacement figure place further is set, quantization step is set, sets up the quantization table that first kind of transfer pair answered according to quantization step that is provided with and displacement figure place according to the number range feature difference of described two kinds of conversion;

The described data that compensate after this first kind of conversion according to the number range feature difference of estimating are: to through the data after first kind of conversion, the quantization table that utilizes described first kind of transfer pair of setting up to answer quantizes.

17. method according to claim 16 is characterized in that, describedly sets up the quantization table that first kind of transfer pair answer and is: it is 2 that point of quantification s corresponding quantitative step-length is set ^{Δ d (s)}* R, wherein R is a constant coefficients value.

18. method according to claim 16, it is characterized in that, this method further comprises: decoding end is set up the inverse quantization table of first conversion in advance according to the number range feature difference, when the data of decoding after first kind of conversion process, inverse quantization table according to described first conversion carries out inverse quantization to the data that receive, and the inverse quantization result is carried out inverse transformation.

19. method according to claim 18 is characterized in that, the described inverse quantization table of setting up first conversion is:

The inverse quantization step-length that point of quantification s correspondence is set is

Wherein, m is that quantization step reduces a half corresponding quantitative point side-play amount, and h and R are a constant coefficients value;

Whenever m the point of quantification of being separated by is set, and corresponding displacement figure place increases by 1.

20., it is characterized in that described at different point of quantification, the estimated image data through the difference DELTA d (s) of number range after two kinds of conversion are respectively according to arbitrary described method in the claim 16 to 19:

At current point of quantification s, for two kinds of conversion, the mean values scope behind the difference computational transformation With

Wherein, Aij is the number range of the capable j row of i pixel data after first kind of conversion of point of quantification s correspondence, Bij is the number range of the capable j row of i pixel data after second kind of conversion in described two kinds of conversion of point of quantification s correspondence, the yardstick of first kind of conversion is n * n, and the yardstick of second kind of conversion is m * m; View data is Avr through the difference DELTA d (s) of number range after two kinds of conversion respectively _B-Avr _A

21. a transform data processing unit is characterized in that, this device comprises the first number range estimation unit, second value scope estimation unit, number range difference unit and conversion compensating unit,

The described first number range estimation unit is used for first kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through first kind of conversion, and offer described number range difference unit;

Described second value scope estimation unit is used for second kind of transformation matrix that conversion is required according to default two kinds of conversion, the number range of estimated image data after through second kind of conversion, and offer described number range difference unit;

Described number range difference unit, be used for according to the number range of passing through respectively after first, second kind conversion, estimate the number range feature difference of two kinds of conversion, and offer described conversion compensating unit, the number range feature difference of described two kinds of conversion be view data respectively through two kinds of conversion and the corresponding difference that quantizes the back number range, or view data is respectively through the difference of number range after two kinds of conversion;

Described conversion compensating unit, be used to receive data to be transformed, and these data are used first kind of conversion in described two kinds of conversion, and according to the described number range feature difference of estimating compensate after this first kind of conversion data so that the number range behind the number range behind first kind of transform and quantization and the second kind of transform and quantization be consistent.

22. device according to claim 21, it is characterized in that, described number range difference unit, the two kinds of conversion quantization step separately that is used for further finding according to view data corresponding quantitative point is estimated the number range feature difference of described two kinds of conversion, and the number range feature difference of described two kinds of conversion is that view data is respectively through two kinds of conversion and the corresponding difference that quantizes the back number range.

23. device according to claim 22 is characterized in that, described conversion compensating unit comprises that adjusting the factor determines subelement, varitron unit, conversion reprocessing subelement and quantize subelement,

The described adjustment factor is determined subelement, and the difference that the number range that provides according to described number range difference unit is provided is determined to adjust the factor, and offers described conversion reprocessing subelement;

Described varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered described conversion reprocessing subelement;

Described conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment factor, and the result is offered described quantification subelement;

Described quantification subelement, the quantization step of answering according to first kind of transfer pair quantizes multiply by the result who adjusts after the factor.

24. device according to claim 22 is characterized in that, the conversion compensating unit comprises point of quantification correction subelement, varitron unit and quantification subelement,

Described point of quantification correction subelement, the difference that the number range that provides according to described number range difference unit is provided is determined the point of quantification side-play amount, and the point of quantification that first kind of transfer pair answered deducted described point of quantification side-play amount, and adjusted result is offered described quantification subelement;

Described varitron unit is used for data to be transformed are used first kind of conversion, and the result after the conversion is offered described quantification subelement;

Described quantification subelement the adjusted point of quantification that provides according to the described point of quantification correction subelement result after to conversion is provided quantizes.

25. device according to claim 22 is characterized in that, described conversion compensating unit comprises to be adjusted the factor and point of quantification correction subelement, varitron unit, conversion reprocessing subelement and quantizes subelement,

The described adjustment factor and point of quantification correction subelement, the difference that the number range that provides according to described number range difference unit is provided determines to adjust the factor and point of quantification side-play amount, and the point of quantification that first kind of transfer pair answered deducted described point of quantification side-play amount, adjusted point of quantification is offered described quantification subelement, the adjustment factor of determining is offered described conversion reprocessing subelement;

Described varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered described conversion reprocessing subelement;

Described conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment factor, and the result is offered described quantification subelement;

Described quantification subelement quantizes multiply by the result who adjusts after the factor according to described adjusted point of quantification corresponding quantitative step-length.

26. device according to claim 22 is characterized in that, described conversion compensating unit comprises to be adjusted coefficient and displacement side-play amount subelement, varitron unit, conversion reprocessing subelement and quantizes subelement,

Described adjustment coefficient and displacement side-play amount subelement, displacement side-play amount when being used for according to the number range feature difference of estimating definite adjustment coefficient and quantification, and offer described conversion reprocessing subelement respectively and quantize subelement, wherein said adjustment coefficient is q '=2 ^{D '}* 2 ⁿ, the displacement side-play amount is n, and wherein n is a positive integer, and D ' is through two kinds of conversion and the corresponding difference that quantizes the back number range;

Described varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered described conversion reprocessing subelement;

Described conversion reprocessing subelement is used for the data after first kind of conversion be multiply by definite adjustment coefficient, and the result is offered described quantification subelement;

Described quantification subelement, the quantization step of answering according to described displacement side-play amount and first kind of transfer pair quantizes multiply by the result who adjusts after the factor.

27. device according to claim 22 is characterized in that, described conversion compensating unit comprises to be adjusted the factor and coefficient side-play amount subelement, varitron unit, conversion reprocessing subelement and quantizes subelement,

The described adjustment factor and coefficient side-play amount subelement, the number range difference that is used for providing according to described number range difference unit is determined to adjust the factor and coefficient side-play amount, and offers described conversion reprocessing subelement;

Described varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered described conversion reprocessing subelement;

Described conversion reprocessing subelement is used for that the data after first kind of conversion be multiply by definite adjustment factor and adds the coefficient side-play amount, and the result is offered described quantification subelement;

Quantize subelement, adjust the factor and add that the result after the coefficient side-play amount quantizes multiply by according to the quantization step that first kind of transfer pair answered.

28. device according to claim 21 is characterized in that, the number range feature difference of two kinds of conversion that described number range difference unit is estimated is that view data is respectively through the difference of number range after two kinds of conversion;

Described conversion compensating unit comprises that quantization table sets up subelement, varitron unit and quantize subelement;

Described quantization table is set up subelement, is used to be provided with the displacement figure place, and according to described number range feature difference quantization step is set, and is that the corresponding quantitative table is set up in first kind of conversion according to quantization step that is provided with and displacement figure place, and offers described quantification subelement;

Described varitron unit is used for data to be transformed are used first kind of conversion, and transformation results is offered described quantification subelement;

Described quantification subelement, the quantization table of answering according to described first kind of transfer pair the result after to first kind of conversion quantize.

Images