TWI280804B - Method for splitting 8x8 DCT into four 4x4 modified DCTS used in AVC/H. 264 - Google Patents

Abstract

The present invention provides that: (1) with the proposed efficient post processes, the existing DCT splitting algorithms still can be applied to complete the MPEG-2-to-AVC/H.264 conversion. (2) For absorbing some operations into quantization step and allowing 16-bit arithmetic implementation, the quantization table is re-defined. With the aid of the invention, an efficient MPEG-2-to-AVC/H.264 transcoder can be realized accordingly, and the de-compression process of the receiver is without any change.

Description

1280804 IX. Description of the invention: [Technical field of invention] This case is a method of splitting two discrete cosine transform blocks, especially a split type II suitable for MPEG-2 and AVC/H.264 discrete cosine core transform A method of discrete cosine transform blocks. [Prior Art] Due to the good compression quality of MPEG-2, including DVD and digital TV/HDTV (DTV/HDTV) promoted by countries in recent years, MPEG_2 is used as the compression standard, which makes MPEG-2 movies widely available. However, the latest video compression standard is AVC/H.264. Unlike MPEG-2, which uses 8x8 blocks for Type 2 Discrete Cosine Transform (Type-Π DCT), AVC/H.264 uses 4x4 blocks. Do an approximate type of two discrete cosine transforms, and with other techniques can achieve better compression. This means that AVC/H.264 can use the amount of information less than MPEG-2 to represent the same video quality, which is a more efficient compression method. To solve the format conversion between MPEG-2 and AVC/H.264, the problems to be solved include motion vector compensation, transform kernel conversion, and intra-picture prediction in the frequency domain. Technique (transform_domainintra prediction), etc., the present invention main 8 6 1280804 proposes an efficient method for the transformation of the discrete cosine transform core. Examining the previous techniques and core transformations for video format conversion, since AVC/H.264 does not use the standard type 2 discrete cosine transform (Type-II DCT), the prior art cannot be applied to MPEG_2 and AVC/H.264. Format conversion between. Academic papers related to the present invention:

Yuh-Jue Chuang and Ja-Ling Wu? uAn Efficient Matrix-Based 2-D DCT Splitter and Merger for SIMD Instructions”, IEICE Trans. On Information and Systems, Vol. E88-D, No· 7, 2005·

Yuh-Jue Chuang and Ja_Ling Wu, “Direct Splitting and Merging of 2-D DCT in the DCT domain'', Digital Signal Processing, Vol· 14, Issue 6, Nov·2004· Yuh-Jue Chuang, Ting-Jian Pan, Ja-Ling Wu, "General Splitting and Merging of 2-D DCT in the DCT Domain", Proceedings of the IEEE International Midwest Symposium on Circuits and Systems, vol 1, pp. 17-20, 2004·

Athanassios N. Skodras, "Direct Transform to Transform Computation," IEEE Trans, on Signal Processing Letters, Vol.6 Νο·8, Aug 1999. R. Dugad and N. Ahuja, ''A Fast Scheme for Image Size Change in The Compressed Domain, 1' IEEE Trans. 1280804 on Circuit and System for Video Technologies, Vol. 11 No. 4, Apr. 2001. Comparison with the present invention: The above paper uses different methods, using the first N/2 points and the latter N The /2-point type 2 discrete cosine transform coefficient (Type_II DCT coefficients) quickly calculates the method of combining the discrete cosine transform coefficients into N points, but the conversion core used in AVC/H.264 is not the discrete cosine transform of the standard type 2 Therefore, the above paper cannot solve the discrete cosine core conversion of MPEG-2 and AVC/H.264. The US patents related to the present invention are described as follows: Patent No. 6 8 6 818 8 Patent Name · Efficient down-scaling of DCT compressed images Comparison with the present invention: the above patent proposes a discrete cosine transform coefficient of type 2 using the first N/2 point and the last N/2 point (Type_n dct coefficient s) Quickly calculate the method of combining the discrete cosine transform coefficients of N points, and use this method to enlarge and reduce the image. The concept is similar to the present invention, but the conversion core used in AVC/H.264 is not standard. The discrete cosine transform of the second type, so the US 6868188 patent and the MpEG-2 and AVC/H.264 discrete cosine transform format conversion proposed by the present invention are not the same. Patent No.: 6577767 8 8 1280804 Patent Name: Transcoding using block -based motion vectors vs. the present invention: This patent is mainly for the conversion between Motion JPEG and MPEG and DV (Digital Video) and MPEG-2, due to the conversion core and Motion JPEG used in AVC/H.264. The DVs are all different, so the Transcoding work proposed in US Pat. No. 6,577,767 cannot be applied to the format conversion between AVC/H.264 and MPEG-2 proposed by the present invention. Patent No.: 6908241 Patent Name: Video transcoder with up-sampling Comparison with the present invention: This patent proposes a method for amplifying the resolution of a movie in a discrete cosine transform domain for an mpeg film. Since the conversion core used in AVC/H.264 is not the discrete cosine transform of the standard type 2, the method proposed in the patent No. 689824 cannot solve the format interchange between MPEG-2 and AVC/H.264 which the present invention focuses on. The problem. Patent No.: 6671332 Patent Name: Video transcoder with spatial resolution reduction. Comparison with the present invention: This patent proposes a method for changing the resolution of a movie in a discrete cosine transform domain, since the conversion core used in AVC/H.264 is not a standard type. Two discrete cosine transforms, therefore,

9 1280804 US Pat. No. 6,671,332 fails to solve the problem of the format interchange between MPEG-2 and AVC/H.264 solved by the present invention. Patent No.:6647061 Patent Name: Video size conversion and transcoding from MPEG-2 to MPEG-4 Comparison with the present invention: This patent aims to resolve the resolution between the film resolution and the film format MPEG-2 and MPEG-4, but The problem of the format interchange between MPEG-2 and AVC/H.264 mentioned in the present invention cannot be solved. For this reason, the applicant has invented the "method of splitting the two-discrete cosine transform block" in view of the lack of the prior art, in order to improve the lack of the above-mentioned conventional means. SUMMARY OF THE INVENTION The main purpose of the present invention is to provide a general post-processing program and a correction for the original method of dividing an 8x8 type two-discrete cosine transform block into four 4x4 type two-discrete cosine transform blocks. Subsequent AVC/H.264 quantization table, in this way, can be applied to the discrete cosine core conversion of MPEG-2 and AVC/H.264. According to the above concept, the present invention provides a method for dividing a type two discrete cosine transform block, the method comprising: receiving an image data; performing variable length decoding and dequantization on the image data to generate a 1280804 8x8 type two discrete cosine transform block; The 8x8 type two discrete cosine transform block is divided into four 4x4 type two discrete cosine transform blocks; a post-processing program A2 is performed on the 4x4 type two discrete cosine transform block to generate four 4χ4 approximate type two discrete cosine transform blocks. Where the post-processing program A? can be expressed as

1 0 0 0 0 0 tan(85.93°) 0 -1 0 0 0 1 0 0 0 1 0 tan(85.93°) 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 _0 0 0 0 0 〇 〇〇〇〇tan(85.93°) 0 1 0 0 0 0 0 0 0 0 0 0 0 -1 1 0 0 tan(85.93°) Quantizes these 4x4 approximations of two discrete cosine transform blocks. As described, the image data is an MPEG-2 format image data, and the steps of quantizing the 4x4 approximation type two discrete cosine transform blocks use the following quantization parameters: (/, j) = j)}[( I m7) 14X A(Qm , /, j) + /2,3+&) »(13 + ρ,)], for (/, j) = {(1,2), (1,4), (3,2), (3,4)} U^{X(/57*)}[(| X{iJ) I 4xA(QM,Uj) + f2l5+^ ) » (15+ 0£)], otherwise m,i) M(QMa) sound (2m,3) fora/) = (0,0),(0,2),(2,0),(2,2) for(/5y) = (1, 1),(1,3),(3,1),(3,3) for(U) = (U),(l,4),(3,2),(3,4) for(U) = (2,l),(0,2),(2,0),(2,2) 1280804 13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233 In the method described, the quantization parameters are modified from quantization parameters defined in the original AVC/H.264 specification.

According to the above concept, the present invention further provides a method for dividing a type two discrete cosine transform block, the method comprising: receiving an image data of an MPEG-2 format; performing variable length decoding and dequantization on the image data to generate an 8x8 type two discrete a cosine transform block; dividing the 8x8 type two discrete cosine transform block into four 4x4 type two discrete cosine transform blocks; performing a post-processing procedure A2 on the 4x4 type two discrete cosine transform block to generate four 4x4 approximation patterns a cosine transform block, wherein the post-processing program A2 can be represented as

1 0 0 0 tan(85.93°) 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 tan(85.93°) 0 0 0 0 0 0 0 0 0 0 0 0 ο ο Ο tan(85.93°) Ο ο ο ο ο ο ο ο ο ο ο — 1 1 ο Ο 1 Ο tan (85.93〇) ; and quantize the 4x4 approximation type two discrete cosine transform blocks. As described, the steps of quantizing the 4x4 approximation type two discrete cosine transform blocks use the following quantization parameters: 12 1280804 sign{X{U 7)}[(1 X(UJ) I 4 x A{QM, U j) + /2u+Qe )»(13 + Qe )], for (/, j) = {(1,2), (1,4), (3,2), (3,4)} sign{ X(z, 7)}[(1 X(/, 7) | 4x A(QM, /, j) + /215+Qe ) »(15 + )], otherwise M(Qm ?0) for (U j ) = (〇?〇)? (〇?2)? (2?0)? (2?2) M(2m,1) for (/,刀=(1,1),(1,3),( 3,1),(3,3) M{Qm ?2) for (/, j) = (1,2), (1,4), (3?2)? (3,4) M{Qm , 3) for (/· j) = (2?1)? (0,2), (2,0), (2?2)

13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233_ As described, these quantization parameters are modified by the quantization parameters established by the original AVC/H.264 specification. Made. According to the above concept, the present invention further provides a method for dividing a type two discrete cosine transform block, the method comprising: receiving an image data of an MPEG-2 format; performing variable length decoding and dequantization on the image data to generate an 8x8 type II a discrete cosine transform block; dividing the 8x8 type two discrete cosine transform block into four 4x4 type two discrete cosine transform blocks; performing a post-processing procedure A2 on the 4x4 type two discrete cosine transform block to generate four 4x4 approximation patterns a discrete cosine transform block, wherein the post-processing program A2 can be represented as

13 1280804; and using the following quantization parameters to quantize the 4x4 approximation type two discrete cosine transform blocks sign{X{U 7)}[(| j) I 4x A{Qm , z, j) + /213+& ) »(13 + )]? for (/, j) = {(1,2), (1,4), (3,2), (3,4)} sign{X(i, 7)} [( 1 X(z, j) | 4xA(QM, i, j) + /215+Qe ) »(15 + )], otherwise for (/, force=(0,0),(0,2),(2 ,0),(2,2) for(/,y) = (l,l),(l,3),(3,l),(3,3) f〇r(U) = (l,2 ),(l,4),(3,2),(3,4) fora/) = (2,l),(0,2),(2,0),(2,2)

10 0 0 0 tan(85.93°) 0 -1 0 0 10 0 1 0 tan(85.93°) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 tan(85.93.) 0 -1 0 0 1 0 0 1 0 tan(85.93

Xq{iJ) = bay m, 3) 13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233 As described, these quantitative parameters are derived from the original AVC/H The quantitative parameters developed by the .264 specification have been modified. [Embodiment] Since the end-user network environment, the computer system 14 1280804, the display device, and the playback tool are different, the content provider often needs to prepare a plurality of different videos for the same movie. The version meets the needs of many users' but it also creates a waste of storage space. Through the instant video format and video transcoding technology, the media content provider only needs to prepare a file, and then can convert the file format or size according to the user's needs, thereby achieving full multimedia storage. Take the purpose of (universal multimedia access). In recent years, due to the popularity of DVDs and HDTVs, the number of compressed movies with [vjpeg,] has been increasing, which has led to the widespread circulation of MPEG-2 movies. Most homes have DVD players or digital TV receivers with MroG-2. The ability of the film. However, the latest video compression standard is AVC/H.264 jointly developed by ITU_T and ISO/IEC, which is the biggest difference between MPEG-2 and 8x8 blocks for Type II II DCT. In fact, AVC/H.264 uses 4x4 squares for "approximate" type two discrete cosine transforms. AVC/H.264 uses an approximate type of two-discrete cosine transform combined with matrix decomposition techniques, which allows the core of discrete cosine transform to be completed using integer arithmetic, plus other techniques developed by AVC/H.264. Compared with MPEG-2, the compression ratio can be improved by 1.5 to 2 times under the same film quality, which means that AVC/H.264 can use less information than MPEG-2 to represent the same movie. Quality is a more efficient compression method. Therefore, we believe that in the future, whether the film format is converted from 8 15 .1280804 MPEG-2 to AVC/H.264 to reduce the file size, or to convert AVC/H.264 movies into the most widely distributed MPEG- 2 videos are absolutely necessary! Please refer to the first figure, which is an architectural diagram of an MPEG-2 to AVC/H.264 transcoder comprising an MPEG-2 decoder π and an AVC/H.264 encoder 12. The MPEG-2 decoder u receives an MPEG-2 format image data, and then the image data is subjected to variable length decoding 111 and dequantization 112 to generate an 8×8 type two discrete cosine transform block, and then for each 8×8 type two discrete The cosine transform block performs inverse discrete cosine transform 113, converts the data to the spatial domain, and transmits it to the AVC/H.264 encoder 12. In order to perform the decoding operation of the image data, the MPEG-2 solution includes executing the operations of the motion compensation 114 and the frame storage 115. When the AVC/H.264 encoder 12 receives the 8χ8 spatial domain data block, the 8x8 square is divided into four 4χ4 sub-blocks, and the four sub-blocks are respectively subjected to a 4×4 approximation type two gas-free chord conversion 121. With quantization 122. In order to perform the operation of the image data, the AVC/H.264 encoder 12 further includes an execution solution 124, a de-discrete cosine transform 125, a frame storage ΐ2β, a motion estimation 127, a motion compensation 128, and a frame prediction 129. And 燏 value editing τ and other operations. As can be seen from the above, the image data is first decompressed by the Mpeg_2 device 11, and then the AVC/H.264 code is crying 'code wU recompressed. Since the AVC/H.264 encoder 12 does not use the Discrete Cosine Transform (Type-IIDCT) of the PM 16 1280804, the previous techniques for video format conversion and core conversion cannot be applied to MPEG-2 to AVC. /H.264 transcoder. The present invention focuses on an efficient method for discrete cosine transform core conversion in MPEG-2 and AVC/H.264 format conversion, avoiding performing inverse discrete cosine transform 113 on the MPEG-2 decoder 11 end and the AVC. The /H.264 encoder 12 performs a forward discrete cosine transform 121. In order to further reduce the amount of calculation, part of the operation is incorporated in the quantization module 122 of the AVC/H.264 encoder 12, so the quantization table required by the quantization module 122 also needs to be modified accordingly. The invention relates to the relationship between the "approximate" type two discrete cosine transform adopted by AVC/H.264 and the standard type two discrete cosine transform, and proposes a post-processing program which is efficient and maintains the quality of the film, combining any "will be an 8x8 type" The method of dividing the two discrete cosine transform blocks into four 4x4 type two discrete cosine transform blocks, that is, the discrete cosine core transform of MPEG-2 to AVC/H.264 can be completed, and vice versa. The two-dimensional case can be regarded as an extension of the one-dimensional case. For the convenience of explanation, the following is a one-dimensional case. Let 厶 be a type 2 discrete cosine transform coefficient of length 8 points, 7/ and 7/ respectively It is a standard type two discrete cosine transform with lengths of 8 points and 4 points, and 7'4 is an approximate type two discrete cosine transform defined by AVC/H.264 with a length of 4 points. Suppose that S is a matrix of 8 X 8 and the method of dividing a length 8-point type two-discrete cosine transform coefficient into two lengths of four-point type 3 17 !28〇8〇4", then S can be expressed by the following formula.

A discrete cosine transform coefficient is expressed as: [Tf] The operation via S' is equivalent to performing an inverse two-discrete cosine transform on a type 2 discrete cosine transform coefficient ' /s ' of length 8 points, [〇

Switch back to the original work area' and then perform a 4-point discrete cosine transform on the first half and the second half, ie [〇. Similarly, fg 0 is another 8x8 matrix, which means "divide a length of a dusty-poly cosine transform vector into two approximate lengths of four points VC/H.264 defined by the approximate type two discrete cosine transform vector" The method 'is defined by the following formula: [τ\]

[Λ] [r8"]-1 (2) In order for the Greeks to use the results of the original $, we will make the post-processing program of the eve of the moon, and the levy of the month, that is, after the processing of ^, The post-treatment of the fruit will be equivalent to the result of the direct permeation treatment. The matrix 可以 can be regarded as 廿'4 ] 0 - \T4r 〇 ' -0 [T\l -0 [τ4γι (3)

18 1280804 Assume /' is a type 2n discrete cosine transform coefficient (Type_n DCT coefficients) block of 8x8, Beckham J "directs an 8x8 type two discrete cosine transform block into four 4x4 AVC/H 264 defined approximations The method of the two-discrete cosine transform block can be expressed as S^rxS^BxSxFx^xg. (4) (5) According to the definition of MPEG_2 and AVC/H.264, 1/2 1/2 1/2 1/2 ' \τπλ= ^C0S^ Visin(f) ~Visin(f) ~ViC0S(f) 4 1/2 -1/2 1/2 1/2 _Visin(f) ~ViC0S(f) Vicos(f) -Visin (f) and 1/2 1/2 1/2 1/2 Therefore V2/5 ViTTo -λ/ΪΤΪΟ -λ/275 1/2 -1/2 -1/2 1/2 _VT7Io -λ/275 V275 - ViTTo 1 0 0 0 0 0 0 cosf+ Visinf 0 s^nf ~Vicosf 0 0 0 vT 0 1 0 0 0 [厶8]= 0 cos Ι V^sinf 0 then cosf + ^/Isinf 0 0 0 Ο 0 0 1 0 0 0 0 0 0 0 0 0 0 0 -\/τ c〇s T + sin |. 0 0 0 0 0 λ/τ cos T ^ sin - 0 0 0 ο (6) ο ο ο ο #cosf+VI Simplify the complexity of matrix 5, we are now 7 Vf %sin (63.43.) and (7) sm lost

19 1280804 • 々βα^63·43. ;), and with the triangle formula, get 4cosf + Wsinf = sin(85.93.) and ^〇^_^ also | = (^(85.93.), after substituting into matrix B, you can get B = according to the process of compression standard, Following the discrete cosine transform is the coefficient deconstruction step. In order to further reduce the computational complexity, we decompose the matrix B and incorporate part of the operation into the quantization table, which is combined with the quantization step. We decompose the matrix B as follows: 1 0 0 0 0 〇sin(85.93°) 〇-cos(85.93°) 0 0 〇1 0 0 〇cos(85.93°) 〇sin(85.93°) 0 0 〇0 0 1 0 〇0 0 0 〇0 0 0 〇0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 sin(85.93°) 〇-c〇s(85.93°) 0 0 1 〇0 cos(85.93°) 〇sin(85.93°) (8) Φ 20 (9) 1280804

5 = for 1 0

A c〇s(85.93°) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 cos(85.93°) 0 0 0 0 0 1 0 0 0 0 0 cos(85.93° 0 0 0 0 0 1 0 0 0 0 0 1 0 0 tan(85.93°) 〇0 〇1 〇0 〇0 〇0 0 ο 0 0 0 0 0 0 -1 0 0 0 1 0 0 0 0 0 tan (85.93°) 0 0 0 0 0 1 0 0 0 0 0 tan(85.93°) 0 0 0 0 0 1 0 0 0 1 0 οοοο οο ο cos(85.93°) Ο ' Ο Ο Ο Ο -1 Ο tan( 85.93°) (1〇) We merge the matrix into the quantization table of AVC/H.264, and the post-processing program 5 can be further simplified to 々, that is, only four multiplication numbers and four addition numbers can be completed. . In order to comply with the 16-bit implementation requirements, the quantitative module related functions and tables defined in the AVC/H.264 specification need to be modified accordingly, including: Quantization formula 4 by j) = sign{ X(/, j)}[(1 X(U j) IA{QU?j) + f2l5+QE )»(\5 + qe)] Changed to 21 (12) (13) 1280804 Quantizes J in Formula 4 M(Qm ?0) for i and j are odds. M{Qm ?1) for i and j are evens M(Qm ,0) otherwise changed to (14) M{Qm ,0) for (/? j) - (0,0), (0,2), (2?0)? (2,2) M{QmX) for (/, force=(14),(1,3),(3,1), (3,3) M(0m,2) for (U) = (1,2),(1,4),(3,2),(3,4) M(2m,3) for (/,y ) = (2,1), (0,2),(2,0),(2,2) The Λ/则 is changed to Μ Μ 13107 5243 8066 11916 4660 7490 10082 4164 6554 9362 3647 5825 8192 3355 5243 7282 2893 4559_ • 13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233 (15) (16) The invention is characterized by: 1. The invention can be combined with any existing "will An 8x8 type two discrete cosine transform block is divided into four The 4x4 type 2 discrete residual 22 1280804 • Chord conversion block method directly divides an 8x8 type two discrete cosine transform block in MPEG-2 into four 4x4 AVC/H.264 defined type two discrete cosines. Convert the block to avoid performing an 8x8 inverse two-fold cosine transform and four 4x4 AVC/H.264 defined approximate two-fold discrete cosine transforms. m Stomach 2. Since the number of bits that the computer can express is fixed, each After a floating-point operation, it is necessary to take close values instead of (rounding), thus causing errors and affecting the quality of the film. The conventional method requires two actions of approximating values, respectively, after performing an 8x8 inverse type two discrete cosine transform. And after four 4x4 AVC/H.264 defined approximate two-fold discrete cosine transform, the present invention directly divides a 8x8 type two-discrete cosine transform block into four 4x4 approximate type two discrete cosine transform blocks. Therefore, it only takes one action to take an approximation, so it is less than the traditional method, so the image quality after format conversion is better than the traditional method. . 3. Modifying the quantization table defined by AVC/H.264 can reduce the amount of computation of format conversion without affecting the decompression process of the file after the end user receives the conversion. 8 23 1280804 Examples and Efficacy of the Invention: The efficacy of the present invention has been empirically simulated and can be illustrated in two aspects. The first aspect is the analysis of the calculated amount, and the second aspect is the analysis of the image quality. 1. Calculation analysis: From the above description, it is shown that the post-processing program 算 (formula (4)) can be completed by four additions and four multiplications, compared with the conventional method s, which requires 272 multiplication numbers and 752 addition numbers, post-processing Program a combines our previous research s (Ref: Yuh-Jue Chuang, Ja_Ling Wu, "Direct Splitting and Merging of 2-D DCT in the DCT domain", Digital Signal Processing, Vol. 14, Issue 6, Nov. 2004) , only 224 multiplication numbers and 464 addition numbers are needed. If you perform computer support single instruction multiple data processing technology (SIMD,

Single Instruction Multiple Data), A (Formula (4)) can be completed by a PMADDWD instruction. In combination with our previous study S, the present invention requires only 6 PMADDWD instructions and 5 PMADD 3 24 1280804 instructions, which is less than the conventional method. PMADDWD instruction and 5 PMADD instructions. 2. Image quality analysis · . First, we use 18 common test images, without considering the subsequent quantization, the test directly divides the 8x8 type two discrete cosine transform block into four 4x4 AVCs by the conventional method and the method of the present invention. The image quality after approximating the two-discrete cosine transform block defined by /H.264, and calculating the PSNR (Peak signal_to-noise ratio) value of the original image. The higher the PSNR value, the closer the original picture is, and the first one is the experimental result. . As can be seen from Table 1, the PSNR values of the method of the present invention are higher than the conventional methods, and the average value is about 3 dB higher. Image PSNR (unquantized) Method of the Invention Traditional Method airfield.bmp 58.922086 55.740291 airplane.bmp 58.875425 55.776364 bridge.bmp 58.829261 55.728460 couple.bmp 58.895136 55.774052 crowd.bmp 58.924472 55.787553 fruits.bmp 62.706741 59.713580 goldhill.bmp 58.907797 55.724269 lax.bmp 58.907797 55.765203 lena.bmp 58.907005 55.770972 25 1280804

Man.bmp peppers.bmp sailboat.bmp splash.bmp statue.bmp tiffany.bmp tool.bmp truck.bmp woman.bmp Avg.PSNR 58.918907 58.942010 58.950805 58.971665 62.893145 58.917318 59.075025 58.945206 58.934827 59.356924 55.710583 55.818183 55.769048 55.741820 59.756727 55.746027 55.922533 55.817016 55.799159 56.214547 We consider the quantified results and calculate the error between the method of the present invention and the conventional method and the original image. The quantization step of AVC/H.264 is from 0 to 51. The larger the quantization step, the higher the efficiency of compression, but the larger the image distortion. We randomly extracted 180,000 8x8 squares from the 18 commonly used test images, and performed two different methods of performing the approximate type two discrete cosine transform defined by AVC/H.264 in two methods. The step size is used to calculate the pixel average error value of each block and the original data. The second figure plots the error value of the conventional method minus the error value obtained by the method of the present invention and the quantization step relationship of AVC/H.264, X. The axis represents the quantization step of AVC/H.264, and the Y axis represents the difference between the two methods. The positive value of Y represents that the error of the method of the present invention is smaller than the error value obtained by the conventional method. 0 8 26 1280804 Quantization level of the second figure The distance relationship is a cycle of six steps. This is because the quantization step of AVC/H.264 is exactly one unit of six. For each additional six levels, the quantization value is doubled. Among them, the original AVC. In the H.264 specification, the design of the multiple of the quantization step size plus two (for example, 2, 8, 14, ..., 50) is particularly well-designed. Therefore, the method of the present invention has effect in these steps and the conventional method. Quite, in addition to the methods produced by the method of the invention The difference is less than the conventional method of generating an error, even when the quantization step is large distance. The superiority of the method of the invention can thus also be demonstrated. In summary, the present invention proposes a general post-processing procedure and a modified method for splitting an 8x8 type two-discrete cosine transform block into four 4x4 type two-discrete cosine transform blocks. AVC/H.264 quantization table, so that it can be applied to the discrete cosine core conversion of MPEG_2 and AVC/H.264, effectively improving the lack of conventional technology, so it has industrial value, and then achieves development. The purpose of this case. This case has been modified by people who are familiar with the art, but it is not intended to be protected by the scope of the patent application. [Simple description of the diagram] The first picture: It is a frame composition of the MPEG-2 to AVC/H.264 transcoder. 8 27 1280804 Second figure: The error value of the conventional method is subtracted from the error value obtained by the method of the present invention and the quantization step distance of AVC/H.264. [Main component symbol description] ll: MPEG-2 decoder 12: AVC/H.264 encoder 111: variable length decoding 112: dequantization 113: inverse discrete cosine transform 114: motion compensation 115: frame storage 221. Discrete Cosine Transform 222··Quantization 223: Entropy Value Encoding 224: Dequantization 225: De-Discrete Cosine Transform 226: Frame Storage 227: Motion Estimation 228: Motion Compensation 229: Interframe Prediction

28

Claims (1)

1280804 日修浅:) is replacing page ten, the scope of patent application: 1 · A split type two discrete cosine inclusion · Receive an image data; variable length decoding and dequantization of the image data, generate an 8x8 type two discrete a cosine transform block; the 8x8 type two-discrete residual 拔 仏 仏 钵 钵 钵 转换 转换 转换 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 分割 ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' = Generate four 4χ4 approximation type two discrete cosine transform squares> The post-processing program h can be table, 1 〇0 ο η, and the method of converting the square, the steps 〇tan(85.93°) 〇— ι 〇0 0 1 0 〇0 1 〇tan(85.93°) 〇0 0 〇0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ο o tan(85.93°) 〇0 0 1 Ο ° 1 〇tan (85.93°); and summer 4x4 approximation type two discrete cosine transform blocks. = The method described in the scope of the patent, wherein the image data is known as an image data in the MPEG_2 format. 3. The method described in claim 1 of the patent scope, the towel quantification, Hai et al. 4x4 approximation type two discrete cosine transform block is the following quantization parameter: (hj) = lsign^x(^ J)} [(1 y) 14 xa(qm 5 /; j)+f2 )>>(13,0 )Ί fr U Chau Deng, side surgery)丨4χΙυ)+/2叫〇>>((]5 + = ^^ W,2 view 29 1280804 For (z·, y·) = (0,0),(0,2),(2,0),(2,2) f〇r(D) = (l,l),(l,3) ,(3,l),(3,3) for (/, 7) = (1,2), (1,4), (3?2)? (3,4) for (/,7)-( 2,1),(0,2), (2,0),(2,2) 13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233 4. The method of claim 3, wherein the quantified parameters are modified from quantitative parameters established by the original AVC/H.264 specification. 5. A method of splitting a type two discrete cosine transform block, the method comprising: receiving an image data in an MPEG-2 format; Μ(βΛ,,〇) M{QMX) m,2) M(Q,n3) M = variable length decoding and dequantization of the image data to generate an 8x8 type two discrete cosine transform block; The type 2 discrete cosine transform block is divided into four 4x4 type two discrete cosine transform blocks; a post-processing program A2 is performed on the 4x4 type two discrete cosine transform block to generate four 4x4 approximate type two discrete cosine transform blocks, wherein the latter The process A2 can be expressed as 30 12808° 4 1 Ο 量化 Quantify the 4x4 approximation type two discrete cosine transform blocks. 6. The method of claim 5, wherein the step of quantizing the 4x4 approximation type two discrete cosine transform blocks uses the following quantization parameters: for (/, y) = (〇?〇χ(〇?2) )? (2,0),(2,2) for(/5y) = (1,1),(1,3), (3,1),(3,3) f〇r(i,/) = (l,2),(l,4)5(3,2),(3,4) f〇r(/?7') = (2,1),(0,2), (2,0 ), (2, 2) 〇 tan (85.93°) And 13107 44933 905 13108' 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 — 7282 25398 512 7233 7 The method of claim 6 wherein the quantitative parameters are The quantitative parameters developed by the original AVC/H.264 specification were modified. 8·- a method for dividing a type two discrete cosine transform block, the step 31 1280804 includes: receiving an image data of an MPEG_2 format; performing variable length decoding and dequantization on the image data, and generating an 8x8 type two discrete cosine transform block; and dividing the product into a 4x4 type by a (four)_(four) two discrete cosine transform block. a discrete cosine transform block; performing a post-processing procedure Α2 on the 4x4 type two-discrete cosine transform block to generate four (four) approximation type two discrete cosine transform blocks, wherein the post-processing program Α2 can be expressed as 0 Π Α ^ . 0 Λ 0 tan(85.93°) 0 一ι 〇〇0 1 〇〇0 1 〇tan(85.93°) 〇〇〇〇〇1 〇0 0 〇〇Ο Ο 〇〇Ο Ο 〇Ο 〇〇Ο 〇〇Ο ο ο ο 1 ο Ο Ο ο tan(85.93°) 〇— ι 〇〇1 Ο 0 1 Ο tan(85.93°) ; and use the following quantization parameters to quantify the 4 χ 4 approximation type two discrete cosine transform blocks j) = \ Sign^x^ I 4x a(Qm 5 hj) + /213+Qe )»(l3-l·QE )], for (/, j) = {(1,2), (1,4), (3 ,2),(3,4)} (/,7)}[(1 ^(/, y) I 4X A(Qm , /, j) + f2^) »(15 + ^)],Otherwise 'Bay m,〇) M{QmX) ' m(qm a) , bay m,3) f Or (/5y) = (〇?〇)?(〇52)? (2,0), (2,2) for (/,7) = (1,1), (1,3),(3, 1), (3,3) f〇r(U) = (l,2),(l,4),(3,2),(3,4) for (/, y) = (2?l) ? (0,2),(2,0), (2,2) 32 1280804 Μ 13107 44933 905 13108 11916 41725 840 11650 10082 36508 735 10485 9362 32450 654 9118 8192 29208 588 8388 7282 25398 512 7233 9·If you apply for a patent The method of clause 8, wherein the quantized parameters are modified from quantization parameters defined by the original AVC/H.264 specification. 33