1310654 九、發明說明： 【發明所屬之技術領域】 本發明係一種數位動態影像快速壓縮方法，尤指一種 應用在H.264/AVC視訊壓縮影像標準的壓縮方法，能夠快 • 速將數位動態影像快速壓縮。 • 【先前技術】 目前由丨TU-T及MPEG兩組織所共同提出之 _ 64/AVC壓細影像標準，在相同壓縮比的條件下，其視 訊品質相較於現有之其它壓縮標準MPEG-2、H.263、 ΜPEG 4等都有相當大的改善。該η_264/AVC壓縮影像標 準係使用許多提昇視訊品質之新技術，例如：可變動區塊 大小移動估測（Variable Block Size Motion Estimation)、 移動估測精確度提升到四分之一像素（M〇t丨〇n Estimat丨〇n1310654 IX. Description of the Invention: [Technical Field] The present invention relates to a digital motion image fast compression method, and more particularly to a compression method applied to the H.264/AVC video compression image standard, which can quickly and digitally display digital motion images. Fast compression. • [Prior Art] At present, the _ 64/AVC compact image standard jointly proposed by TU-T and MPEG organizations, under the same compression ratio, its video quality is comparable to other existing compression standards MPEG-2. , H.263, ΜPEG 4, etc. have a considerable improvement. The η_264/AVC compressed image standard uses many new techniques to improve video quality, such as Variable Block Size Motion Estimation and motion estimation accuracy up to a quarter pixel (M〇) T丨〇n Estimat丨〇n
Accuracy Up To Quarter Sample)，以及位元率壓縮比最 佳化（Rate.Distortion Optimization)等。 _ H.264/AVC視訊壓縮標準係可由各種高低階之程式語 言C、C + +、java、VB或是C#組合語言予以實現之，或 是由純硬體的方式加以製作’亦可由中央處理器（Center Processing Unit，CPU)組合部份硬體構成一軟硬體來完 成，此外，也能直接以晶片系統（System〇n chip，SoC) 予以完成。其執行平台可以是個人電腦（Pers〇na丨Accuracy Up To Quarter Sample), as well as Rate. Distortion Optimization. _ H.264/AVC video compression standard can be implemented by a variety of high- and low-level programming languages C, C + +, java, VB or C# combined language, or produced by pure hardware. The central processing unit (CPU) combines a part of the hardware to form a hardware and a hard body, and can also be directly implemented by a system system (SoC). Its execution platform can be a personal computer (Pers〇na丨
Computer，PC )、各種嵌入式系統（如手機，pda，數位 相機等）’由硬體（如ASIC)或是由上述各種軟硬體以各種 形式組合而構成之系統等。 1310654 請參閱第四圖所示，係為H.264/AVC壓縮影像標 —執行壓縮的方塊圖，主要包含有· -預測模組（3 0) ’係包含有—外部預測單 1) ' -㈣㈣單元（32)及—關模式決定單 ㈧’其中該外部預測單元（31)係、分別取得 行壓縮的畫面⑽巨區塊(係指-張畫面中以16像素乘16 像素大小的影像區塊）Λ前幾張（卜16張的範圍 選定）的晝面根據各外部預測模式，執行各模式 定的搜尋及運算’㈣部預測單元（32) 進行壓縮的畫面€的巨區塊，根據各内部預測模式，執; 各模式所規定的運算H外部關單元（Η)盘 ㈣單元（32”Ρ會計算出的該預測單元中各項預測模 式所需的代價（cost)，之後再將今 、、 争m Q W 便再將錢^輪出至預測模式 厭、疋早兀（33) ’供該預測模式決定單α(33)依昭 =縮::：決定出影響影像品質及位元率最低的議 式，並依該預測模式計算出預測巨區塊； β 一 ：換編碼單元（4 〇)，其包含有一轉換器（川' -里化器（4 2 ) &一熵編碼器（4 3 ) (“）係透過-減法器"4)連接至該預== 測模式決定單元（33)及目前晝面 式:T*(33)輸出的預測巨區塊與目前書面的巨區 塊相減後取得剩餘資料，再將該剩餘資料予以轉換、量化 及熵編碼，而產生壓縮後的位元流； 還原早几（5 Q )，係連接至該轉換單元（4 〇 ) 4 1310654 之量化器（42)的輸出端，其包含有一反轉換器（51) 及一反量化器（5 2 )，其中該反轉化器（5 1 )係透過 一加法器（5 3 )連接至該預測模式決定單元（3 3 )的 輸出端，將經過轉換及量化的剩餘資料予以反量化及反轉 換後，再將預測晝面與剩餘資料相加，以還原出目前的晝 面，並輸出至該内部預測單元，或進一步透過一方塊效應 濾波器（5 4 )濾波後輸出。Computer, PC), various embedded systems (e.g., mobile phones, pdas, digital cameras, etc.) are systems formed by hardware (e.g., ASIC) or a combination of various types of hardware and software described above in various forms. 1310654 Please refer to the fourth figure, which is the block diagram of H.264/AVC compressed image mark-execution compression, which mainly includes · - prediction module (3 0) 'includes - external prediction list 1) ' - (4) (4) Unit (32) and - mode mode decision list (8) 'where the external prediction unit (31) is the picture (10) macro block that obtains the line compression respectively (refers to the image area of 16 pixels by 16 pixels in the picture - picture) Block) The first few sheets (the range of the 16 sheets are selected) are executed according to the respective external prediction modes, and the search for each mode and the calculation of the '(four) part prediction unit (32) to compress the huge block of the screen are performed, according to Each internal prediction mode, the operation of each mode, the operation of the H external closing unit (Η) disk (four) unit (32" Ρ calculated cost of the prediction mode of the prediction unit in the prediction unit, and then will be ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, The lowest formula, and calculate the predicted giant block according to the prediction mode; β 一: a coding unit (4 〇), which includes a converter (Chuan '-Liuizer (4 2 ) & an entropy encoder (4 3 ) (") is a pass-subtractor " 4) connected to the pre- == The measurement mode decision unit (33) and the current facet type: T*(33) output predicted block is subtracted from the current written giant block to obtain the remaining data, and then the remaining data is converted and quantized. Entropy coding, and generating a compressed bit stream; reducing the early (5 Q ), is connected to the output of the quantizer (42) of the conversion unit (4 〇) 4 1310654, which includes a reverse converter (51) And an inverse quantizer (52), wherein the inverter (5 1 ) is coupled to the output of the prediction mode decision unit (33) through an adder (53), which is converted and quantized After the remaining data is dequantized and inversely converted, the predicted surface and the remaining data are added to restore the current surface and output to the internal prediction unit, or further through a block effect filter (5 4 ) Filtered output.
上述内、外部預測單元係依照H.264/AVC壓縮影像標 準所提出下表三的預測模式分類表進行檢驗程序。 畫面種類(Frame Type) 可使用之預測種類 內部 16x16 預測模式(Intra 16x16 Prediction Mode) I 畫面(I Frame) 內部 4x4 預測模式(Intra 4x4 Prediction Mode) 內部 16x16 預測模式(Intra 16x16 Prediction Mode) 內部 4x4 預測模式(Intra 4·χ4 Prediction Mode) 外部 16x 16 預測模式(Inter 16x 16 Prediction Mode) 外部 16x8 預測模式(Inter 16x8 Prediction Mode) 外部 8x16 預測模式(Inter 8x16 Prediction Mode) P 畫面(P Frame) 外部 8x8 預測模式(Inter 8x8 Prediction Mode) 外部 8χ4 預測模式(Inter 8x4 Prediction Mode)* 外部 4x8 預測模式(Inter 4x8 Prediction Mode)* 外部 4x4 預測模式(Inter 4x4 Prediction Mode)* 省略模式(Skip Mode) 表三中的I晝面（Intra Frame)及P晝面（Inter Frame) 係為一般視訊壓縮晝面的兩種分類，其中丨畫面是指整個 5 1310654 視訊壓縮流程中，佔壓縮時間最少但壓縮比最低之畫面， 其僅利用該畫面自身周圍的訊息來壓縮，故失真較少；而 P晝面則是利用先前壓縮過的丨畫面或是p畫面來進行壓 縮，故可達到較高的壓縮比。該p晝面所使用的原理是在 視訊當中，前晝面與後畫面之間有會相當大的機率是有相 互關聯的，因此如能找出其移動關係，僅壓縮移動向量 (Motion Vector)和差異值（Resjdua| Va|ues)，不必全張書 面進行壓縮即能大幅地提升壓縮比。 請配合參閱第五圖所示，係為H.264/AVC壓縮影像標 準對其中I畫面及p畫面進行壓縮的流程： 首先係針對目前欲壓縮的畫面進行判斷（6〇); 若為I畫面，則配合表三中i晝面種類所包含兩個預 測模式，分別進行檢驗（61)，以挑選出其中最低代價的 預測模式（63)所得出的預測結果，再進行壓縮運算（64); 及 若為P畫面，則配合表三中P晝面種類所包含的十個 預測模式逐-進行檢驗（62)，以挑選出其中最低代價的 預測模式（63 )所得出的預測結果’進行壓縮運算…）。 就整體壓縮程序來說，H.264/AVC壓縮影像標準不論 在視訊品質以及壓縮效率上都有明顯地改進，若再使用位 :率失真比最佳化，即能再進一步的提高壓縮品質（此指 前述之梘訊品質以及壓縮效率），但實際上目前h.264/avc 壓縮影像標準欲達到此般的壓縮品質，該所需要的計算量 (Computation Complexity)相當龐大。導致 H 264/M/c 1310654 壓縮影像標準難以應用 夂合武的移動获 D㈣·叫上，甚至連_般個人電 =(叫He 時《。㈣究其中㈣不難發現，到即 標準的壓縮流程大部分的計算時心影像 測模式的決定，亦即執 ’、費於各個巨區塊預 時間較長，特別是以外部預測使/預：、一 Esti_i〇n，ME)演算更花時：：（Μ_Π 估測的演算法被提出以減 則已有許多移動 以達到即時壓縮的目的。 、之计斤置，斯 同樣地也有相告數量的_ Η古測也有類似的情形， 是針對於單—個巨區塊如何 速，但其均 縮短檢驗時間，#目# 凋模式進行檢驗時， 要。 目“效均非理想，有進-步改進之必 【發明内容】 單巨m的主要發明目的係提供—種完全跳脫研奸fi _AVC璧缩…：間檢驗的演算，而是將目前 分類，令各^所提出的預測種類表加以統計、 刀頰7各早巨區塊不必—— 驗，能加速進入屡缩….的預測模式進行檢 {進入£縮運异的時間，是以 縮短内、外部關的決定時間，令H =了有效地 準的整體壓縮時間大幅縮短。…⑽崎縮影像標 方法目的所使用的主要技術手段係令該加速I縮 先對H.264/AVC I缩影像標準中所提出預測模式分類 1310654 表進行分類 至少三個種_照所有預測模式運算資料量大小，分成 第—牛、，如此各種類包含有少數的預測模式； 擇第—種係ΐ預定複數種類的預設模式分類表中選 式檢驗，進並將單巨區塊逐—與第—種類的少數預測模 模式；龙中:所規定之代價估算，再決定合適的第一預測 準所提預設模式分類表係對H.264/AV4縮影像標 算資料量大_換式分類表中的戶畫面預測模式，依照運 類包含少數預測模式； “為二大種類，令各種 預定依照第一步嶋程序中產生的參數，自 “ /員的預设模式分類表中選擇第二種類；及 若曰判斷第-種類預測模式是否等於第二種類預測模式， :’則以第-步驟的已計算出此預測模式的代價結果； 則將單巨區塊逐一與第二種類該少數預測模式進行 :驗，進行此模式所規定之代價估算，再決定合適的第二 測桓式，求後比較第-預測模式與第二預測模式的代價 :果、，最小代償者為最後預測模式，再以此模式進行塵縮 、’扁*5馬運算。 由上述步驟說明可知’本發明對—張目前數位影像書 产面進行壓縮’係將每張數位影像晝面的各巨區塊重覆進行 第-、第二步驟的壓縮，即能完成壓縮。由於本發明第—' 第:步驟係主要將習知單巨區塊與所有預測模式進行檢驗 運异，以統計分析方式找出適合的預測模式，是以，本發 明並無需將巨區塊逐一與所有預測模式進檢驗，而是藉= 1310654 _ 減少檢驗之模式達成加速壓縮之目的。 由於但事實上若能在執行眾多内部預測模式或外部預 測模式之前先行評估其發生之可能性並省去部分模式，即 可先一步的將所需壓縮時間大幅減少。 ' 【實施方式】 本發明係主要針對H.264/AVC壓縮影像標準中的模式 檢驗及模式決策兩部份進行改良，令巨區塊在進行決定適 當預測模式的時間得以有效縮減，進而大幅改善壓縮時 ^間。 請參閱下表一所示，係為本發明對於H.264/AVC壓縮 影像標準中所提出預測模式分類表，進一步依照運算資料 量大小進行分類，而供本發明壓縮方法使用的預定複數種 類的預設模式分類表，表一為本發明分類表的第一較佳實 施例，其係分成四個種類（大、小、省略及内部）：The above internal and external prediction units perform the verification procedure in accordance with the prediction mode classification table of Table 3 below proposed by the H.264/AVC compressed image standard. Frame Type Available Predicted Type Internal 16x16 Prediction Mode (Intra 16x16 Prediction Mode) I Picture (I Frame) Internal 4x4 Prediction Mode (Intra 4x4 Prediction Mode) Internal 16x16 Prediction Mode (Intra 16x16 Prediction Mode) Internal 4x4 Prediction Mode (Intra 4·χ4 Prediction Mode) External 16x 16 Prediction Mode (Inter 16x 16 Prediction Mode) External 16x8 Prediction Mode (Inter 16x8 Prediction Mode) External 8x16 Prediction Mode (Inter 8x16 Prediction Mode) P Frame (P Frame) External 8x8 Prediction Mode (Inter 8x8 Prediction Mode) External 8χ4 Prediction Mode (Inter 8x4 Prediction Mode)* External 4x8 Prediction Mode* Inter 4x4 Prediction Mode* Interference Mode (Skip Mode) Intra Frame and Inter Frame are two categories of general video compression, and the frame is the image with the lowest compression time but the lowest compression ratio in the entire 5 1310654 video compression process. , it only uses the information around the picture itself to compress, so the distortion is better ; Day of P side is using the previously compressed or p Shu picture to picture compression, it can achieve a higher compression ratio. The principle used in this p-plane is that in video, there is a considerable probability that there is a high probability between the front and back pictures, so if you can find out the movement relationship, only compress the motion vector (Motion Vector). And the difference value (Resjdua| Va|ues), which can greatly increase the compression ratio without having to compress it in full. Please refer to the fifth figure for the H.264/AVC compressed image standard to compress the I picture and p picture: First, judge the picture to be compressed (6〇); Then, in accordance with the two prediction modes included in the table i in the third table, respectively, the test (61) is performed to select the prediction result obtained by the lowest cost prediction mode (63), and then the compression operation is performed (64); And if it is a P picture, it is tested by the ten prediction modes included in the P face type in Table 3 (62), and the prediction result obtained by selecting the lowest cost prediction mode (63) is compressed. Operation...). As far as the overall compression program is concerned, the H.264/AVC compressed image standard has been significantly improved in both video quality and compression efficiency. If the bit: rate distortion ratio is optimized, the compression quality can be further improved. This refers to the aforementioned quality and compression efficiency), but in fact the current h.264/avc compressed image standard wants to achieve such compression quality, and the required calculation complexity (Computation Complexity) is quite large. Leading to the H 264/M/c 1310654 compressed image standard is difficult to apply 夂 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的 的Most of the calculations in the calculation of the heart image measurement mode, that is, the implementation of the fee, the pre-exposure time of each macro block, especially in the external prediction / pre:: an Esti_i〇n, ME) calculations more time ::(Μ_Π The estimated algorithm is proposed to reduce the existing movements to achieve the purpose of instant compression. The calculation of the amount, the same number of _ _ Gu Gu also has a similar situation, is aimed at In the case of a single block, how fast is it, but it all shortens the inspection time, when the #目# mode is tested, it is required. The effect is not ideal, there is a step-by-step improvement [invention content] single giant m The main purpose of the invention is to provide a kind of calculation for the complete test of the traitor fi _AVC contraction, but the classification of the current classification, so that the prediction type table proposed by each ^ is counted, and the early blocks of the cheeks are not necessary. —— Test, can accelerate the prediction mode into the contraction Checking the time to enter the contract is to shorten the decision time of the internal and external customs, so that the overall compression time of H = effective and accurate is greatly shortened. (10) The main technical means used for the purpose of the image reduction method The acceleration I contract is first classified into the prediction mode classification 1310654 table proposed in the H.264/AVC I thumbnail image standard. At least three kinds of data are calculated according to the size of all prediction modes, and are classified into the first cow, and thus various classes. Contains a small number of prediction modes; Select the first-type system to pre-determine the plural types of the default mode classification table, select the test, enter the single-block block by- and the first-type minority prediction mode; The estimated cost estimate, and then determine the appropriate first forecasting standard. The default mode classification table is for the H.264/AV4 thumbnail image data. The household picture prediction mode in the classification table is included in the class. a small number of prediction modes; "For the two major categories, the various schedules are selected according to the parameters generated in the first step of the procedure, and the second category is selected from the /person's preset pattern classification table; and if the first type prediction mode is judged Whether it is equal to the second type of prediction mode, : ' then the cost result of the prediction mode has been calculated by the first step; then the single macro block is performed one by one with the second type of the minority prediction mode: the test is performed according to the mode Estimate the cost, then determine the appropriate second measurement method, and compare the cost of the first prediction mode and the second prediction mode: fruit, the minimum compensator is the final prediction mode, and then the mode is used for dust reduction, 'flat *5 horse operation. From the above description, it can be seen that 'the present invention compresses the current digital video book production face' by repeating the first and second steps of each macro block of each digital image. That is, the compression can be completed. Since the first step of the present invention mainly tests the conventional single block and all the prediction modes, and finds a suitable prediction mode by statistical analysis, the present invention does not need to The macroblocks are tested one by one with all prediction modes, but by the =1310654 _ reduction test mode to achieve accelerated compression. However, in fact, if you can evaluate the possibility of occurrence and save some of the modes before executing many internal prediction modes or external prediction modes, you can reduce the required compression time step by step. [Embodiment] The present invention mainly improves the mode check and mode decision in the H.264/AVC compressed image standard, so that the time when the macro block determines the appropriate prediction mode is effectively reduced, thereby greatly improving When compressing ^. Referring to Table 1 below, the prediction mode classification table proposed in the H.264/AVC compressed image standard of the present invention is further classified according to the amount of operation data, and is used for the predetermined plurality of types of the compression method of the present invention. The preset mode classification table, Table 1 is a first preferred embodiment of the classification table of the present invention, which is divided into four categories (large, small, omitted, and internal):
種類(Type) 包含模式(Modes) 外部 Ιόχΐ6 預測(Inter 10x16 Prediction Mode) 大(Big) 外部 16又8 預測(Inter Ιόχδ Prediction Mode) 外部 δχΐ6 預測(Inter δχΐ6 Prediction Mode) 外部 8x8 預測模式(Inter 8x8 Prediction Mode) 外部 8x4·預測模式(Inter 8x4 Prediction Mode)* 小(Small) 夕'^部 4x8 預測模式(Inter 4x8 Prediction Mode)* 夕f部 4x4 預測模式(Inter 4x4 Prediction Mode)* 省略(Skip) 省略模式(Skip Mode) 內部 16x16 預測模式(Intra 16x16 Prediction Mode) 內部(Intra) 9 1310654 內部 4x4 預測模式(Intra 4x4 Prediction Mode) 此外，如下表二亦可進一步對上表的大種類再分出另 一個中種類，共有五個種類，為本發明分類表的第二較佳 實施例： 種類(Type) 包含模式(Modes) 大(Big) 外部 16x16 預測(Inter 16x16 Prediction Mode) 外部 Ιόχ8 預測(Inter Ιόχ8 Prediction Mode) 中(Medium) 外部 8x16 預測(Inter 8xl6 Prediction Mode) 外部 8x8 預測模式(Inter 8x8 Prediction Mode) 夕'^部 8x4 預測模式(Inter 8x4 Prediction Mode)* 小(Small) 外部 4x8 預測模式(Inter 4x8 Prediction Mode)* 外部 4x4 預測模式(Inter 4x4 Prediction Mode)* 省略(Skip) 省略模式(Skip Mode) 內部 16x16 預測模式(Intra 16x16 Prediction Mode) 內部(Intra) 內部 4x4 預測模式(Intra 4x4 Prediction Mode)Type Contains Modes External Ιόχΐ6 Prediction (Inter 10x16 Prediction Mode) Large (Big) External 16 and 8 Prediction (Inter Ιόχδ Prediction Mode) External δχΐ6 Prediction (Inter δχΐ6 Prediction Mode) External 8x8 Prediction Mode (Inter 8x8 Prediction Mode) External 8x4 · Prediction Mode (Inter 8x4 Prediction Mode) * Small (Small) 夕 '^ 4x8 Prediction Mode (Inter 4x8 Prediction Mode) * 夕 f 4x4 Prediction Mode (Inter 4x4 Prediction Mode) * omitted (Skip) omitted Mode (Skip Mode) Internal 16x16 Prediction Mode (Intra 16x16 Prediction Mode) Internal (Intra) 9 1310654 Internal 4x4 Prediction Mode (Intra 4x4 Prediction Mode) In addition, Table 2 below can further divide the larger category of the above table. There are five categories in the category, which are the second preferred embodiment of the classification table of the present invention: Type (Mode) Mode (Big) External 16x16 Prediction (Inter 16x16 Prediction Mode) External Ιόχ8 Prediction (Inter Ιόχ8 Prediction Mode) Medium 8x16 Prediction Mode (Inter 8xl6 Prediction Mode) External 8x8 Prediction Mode (Inte) r 8x8 Prediction Mode) Inter 8x4 Prediction Mode (Inter 8x4 Prediction Mode) * Small (Small) External 4x8 Prediction Mode (Inter 4x4 Prediction Mode) * External 4x4 Prediction Mode * Intermitted (Skip) omitted Skip Mode Internal 16x16 Prediction Mode (Intra 16x16 Prediction Mode) Internal (Intra) Internal 4x4 Prediction Mode (Intra 4x4 Prediction Mode)
本發明係主要以兩道統計比對步驟，以決定出各巨區 塊的預測模式，請參閱第一圖所示，係為本發明方法的執 行流程，其包含有： 判斷晝面形式（10 )，若為丨晝面或緊鄰I晝面的P 晝面，則執行原始H.264/AVC壓縮影像標準中I晝面以及 原始P晝面的檢驗（1 1 )、決定模式（1 2 )及壓縮運算（1 3 ) 程序；若為其他P晝面則執行以下步驟； 第一步驟（20 ):係自預定複數種類的預設模式分類 表中選擇第一種類（201 )，並將單巨區塊逐一與第一種 10 1310654 類的少數預測模式進行檢驗（2〇2 )，再決定適合的第一 預測模式（203 );又，第一步驟進行檢驗、模式運算的 過程中會產生的複數參數係包含有絕對差異之總和 Absolute Diiierence)、絕對差異平方之總和（Sum Squared Absolute Difference)、移動向量編碼後之位元 數（MV Bits)和剩餘巨區塊編碼（Residual Macr〇M〇ck出⑷ 後之位元數等其他在運算過程中產生的資訊； 第二步驟（21):依照第一步驟.壓縮程序中產生的參 數，自預定複數種類的預設模式分類表中選擇第二種類 (211 );及 ' 判斷第種類預測模式是否等於第二種類預測模3 (212)，若是，則以第一步驟的已估算出最後所要使用 的預冽杈式，該預測模式則為後續壓縮運算所使用（23); 右否’則將單巨區塊逐_與第二種類該少數預測模式進行 檢驗（213) ’再決定適合的第二預測模式（214);比較 孩第—預測模式與第一預測模式所需的壓縮代價（22)，最 小代價的預龍式為㈣壓縮運算使用之模式（23)。 ，上述第一步驟的選擇第-種類（201) &方法，請參 閱第二圖所示，係包含有·· 、知複數個在目蝻單巨區塊之前已壓縮巨區塊之預測 二（。⑷°月配合參閲第三HA、B所示，複數先前 區塊㈣®係指於本晝面（M2)中與目前進行麼 :巨區塊（X)的前四個已被壓縮的巨區塊“〜幻，以及 j張晝面（M1 )中以對應目前巨區塊位置為中心的九個 1310654 巨區塊（AH),如此共 絲·^ t = η·. 丁 —们已壓縮的巨區塊； 计複數已壓縮巨區塊的預測 模式分㈣，W % 制H，以及對照該預設 第一種類（觸）。 、气的種類，該種類即為 模式進表及表-’略種類均為獨立的模式，且此省略 、工進仃檢驗所需之計算複雜度 檢驗機率，上述第—…一： ^此省略模式的 類en ^驟進一步包含有判斷選擇的第一種 員疋否為省略種類C 204 )，若否，則古4,. 的檢龄f one、、 、]直接進行省略模式 第牛 ，並取得賓略模式的參數（ 205a)，進入 若是，則以上一步驟決定種類所計算出的參數 進入下第二步驟。 紙 又，上述第二步驟選擇第二種類方法（2彳1)包含有· 準備-轉換機率表（211a)，係對數個標準視訊資料 0㈣化參數下’給定第„步驟所選用之種類以及其產 斗H數後’統計第_步驟選用的種類再轉換至其他各種 類的機率；是以，該轉換機率表的產生方式是先針對數個 視訊進行塵縮，而後統計參數在第一步驟選定之檢驗種類 的條件下，轉換至其他種類之機帛’則假設在第一步當中 所選定之檢驗種類為PreTpi，其所使用之參數為\、χ2、1、 最後轉換至種類丁之機率即可定義為下式：2 PT^P(T\VxeTPl,xvx2,...iXn)； 如此，種類τ可能為任一種類包含PreTp]。根據該式， 統計各個視訊壓縮後之結果，來完成該表格，而每一可能 機率種類都須達一定數量後才可適用於預測判斷。即’以 12 1310654 1310654 表一所不的預測模式分類方式（其他各種分類法亦屬於本發 明之範圍）的任何一個種類，在PreTpi所選取之種類條件 下都有一機率跳至其他或自己本身，&纟某特定參數x]、 Χ2、...、Xni情況下，共計有4(跳至種類）χ4(種= 個機率，如下表所击:The present invention mainly uses two statistical comparison steps to determine the prediction mode of each macro block. Please refer to the first figure, which is an execution flow of the method of the present invention, which includes: judging the face form (10) ), if it is a facet or a P face immediately adjacent to the I face, perform the test of the I face and the original P face in the original H.264/AVC compressed image standard (1 1 ), and determine the mode (1 2 ) And the compression operation (1 3 ) program; if it is another P plane, the following steps are performed; the first step (20): selecting the first category (201) from the predetermined plurality of preset pattern classification tables, and The macroblocks are tested one by one with a few prediction modes of the first type 10 1310654 (2〇2), and then the appropriate first prediction mode (203) is determined; in addition, the first step is performed during the test and the mode operation. The plural parameters include Absolute Diiierence), Sum Squared Absolute Difference, MV Bits after moving vector coding, and Residual Macr〇M〇 Number of bits after ck out (4) And other information generated during the operation; second step (21): selecting the second category (211) from the predetermined plurality of preset pattern classification tables according to the parameters generated in the first step of the compression program; and Determining whether the first type prediction mode is equal to the second type prediction mode 3 (212), and if so, estimating the last used preamble in the first step, the prediction mode is used for subsequent compression operations (23) ; right no' then test the single macroblock _ with the second type of the minority prediction mode (213) 'determine the appropriate second prediction mode (214); compare the child-prediction mode with the first prediction mode The required compression cost (22), the minimum cost of the pre-dragon type is (four) compression operation mode (23). The first-step selection of the first-type (201) & method, please refer to the second figure, The system contains the predictions of the compressed macroblocks before the single block is seen. (4)° month with reference to the third HA, B, the plural previous block (four)® refers to What is the current and the current (M2): giant block (X) The first four compressed giant blocks "~ illusion, and j Zhang 昼 (M1) in the nine 1310654 giant blocks (AH) centered on the current giant block position, so collinear ^ t = η·. Ding—the giant blocks that have been compressed; the prediction mode of the complex massive compressed block (4), W % for H, and the preset first type (touch). That is, the mode entry table and the table--slightly all types are independent modes, and the omission of the computational complexity check probability required for the work-inspection test, the above-mentioned first: ^ This omission mode class en further includes The first member who judges the choice is omitting the category C 204 ), and if not, the age of the old 4,. f one, , , and ] directly omits the mode of the first cow, and obtains the parameter of the objective mode ( 205a If yes, the above step determines the parameter calculated by the category and proceeds to the next second step. Paper, the second step of the second step of selecting the second type method (2彳1) includes a preparation-conversion probability table (211a), which is a type of standard video data 0 (four) parameter selected under the given step „ After the production of the H number, the statistics of the type selected in the _ step are converted to other various types of probability; that is, the conversion probability table is generated by first performing dust reduction for several videos, and then the statistical parameters are in the first step. Under the condition of the selected inspection type, the conversion to other types of machine 帛' assumes that the type of inspection selected in the first step is PreTpi, and the parameters used are \, χ 2, 1, and finally convert to the type din. It can be defined as follows: 2 PT^P(T\VxeTPl, xvx2,...iXn); Thus, the category τ may contain PreTp for any kind. According to this formula, the result of each video compression is counted. Completion of the form, and each possible probability type must be up to a certain amount before it can be applied to the forecasting judgment. That is, the classification mode of the prediction mode is not included in Table 13 (the other various classification methods are also within the scope of the present invention) Any one of the categories, under the type of condition selected by PreTpi, has a chance to jump to other or itself, & 纟 a specific parameter x], Χ 2, ..., Xni, a total of 4 (skip to category) χ 4 (species = probability, as shown in the following table:
-~-- 因此，若在不同參數 ^ - 對情形下會有不同之數值，仍2亦屬本:;之下’機率表相 特定參數…、·,·、χ二!屬本發明之範圍内。在 則機率相對變大，但仍亦屬：配合表二的分類方式， 丌屬本發明之範圍内. 以第-步驟決定的第一種類對照機率轉換 子應第-種類的最高機率種類（川b) ·、以查出 判斷在機率轉換表中的第—步驟_，日 最有可能轉換的新種類（211c). '疋否對應單一 若否，則執行一步驟；決定二個 種類⑺⑷，係可以類神經網路進 為第二 :路係以第-步驟所產生的複數參 以:：類神缓 力，實際應用時’即可將第 “丨練分類能 類神經網路中，以決定出生的複數參數輪入 …現此-目的的 1310654 類神經網路係包含有倒傳遞類神經網路(Back Propagatlon Neural Network，βΡΝΝ)或半徑式函數網路㈤^如仏 ^山⑽，咖）等，或者其它所有使用人工訓練 或機器學習應用於模式區分之類神經網路亦都可使用也包 含在本發明之範圍内； 若是，則令最高機率種類為第二種類預測模式（211句， 及判斷與第—種類預_式，是否為同-種類（212)，若是， ，直接將第-預測模式為最後預測模式，並為後續屋縮運 r使用(23) ’右否’則將單巨區塊逐—與第二種類該少數 預測模式進行檢驗（213)，再決定適合的第二預測模式 (214 )，及比較该第二預測模式與第一預測模式，較小 壓縮代價者為最後決定之預測模式（22) ’最後將此預測模 式為後續壓縮運算使用（23)。 由上述說明可知，本發明不必將每個巨區塊與所有預 測模式進打檢驗後求出最小代價的預測模式，係主要分開 兩個步驟，每個步驟僅與其中—小種類的數個預測模式進 行檢驗，故就在預測決定時間上明顯縮短許多，對於整個 壓縮時間是能夠被大幅改善的。 【圖式簡單說明】 第一圖：係本發明應用於H.264/AVC壓縮影像標準的 壓縮流程圖。 第二圖：係第一圖的詳細流程圖。 第三圖A ' B :係分別為本發明進行壓縮時所用的前 一張晝面及目前晝面的巨區塊。 1310654 第四圖：係壓縮標準的方塊圖。 第五圖：係既有Η.264/AVC壓縮影像標準的壓縮流程 【主要元件符號說明】 (3 0 )預測模組 （3 1 )外部預測單元 (3 2 )内部預測早元（3 3 )模式決定早元 (4 0 )轉換編碼單元（4 1 )轉換器-~-- Therefore, if there are different values in the case of different parameters ^ -, then 2 is also the present:; below the probability parameter specific parameters ..., ·, ·, χ二! belongs to the scope of the present invention Inside. In this case, the probability is relatively large, but it is still: in accordance with the classification method of Table 2, it belongs to the scope of the present invention. The first type of the probability conversion converter determined by the first step should be the highest probability type of the first type. b) · To find out the new type (211c) that is most likely to be converted in the first step _ in the probability conversion table. '疋No corresponds to a single if no, then performs one step; determines two types (7)(4), The system can be used as the second type of neural network: the path is generated by the plural steps of the first step:: the kind of god-like force, in the actual application, the first can be classified into the energy-like neural network. Decide the birth of the plural parameter wheel... The current 1310654 class of neural network system includes the Back Propagatlon Neural Network (βΡΝΝ) or the radius function network (5)^如仏^山(10), ), or any other neural network that uses manual training or machine learning for pattern differentiation, and is also included in the scope of the present invention; if so, the highest probability type is the second type of prediction mode (211 sentences) And judgment And the first-type pre-type, whether it is the same-type (212), and if so, directly use the first-prediction mode as the final prediction mode, and use for the subsequent house reduction r (23) 'right no' will be single The block is tested with the second type of the minority prediction mode (213), and then the appropriate second prediction mode (214) is determined, and the second prediction mode and the first prediction mode are compared, and the smaller compression cost is the last. The prediction mode of decision (22) 'The final prediction mode is used for subsequent compression operations (23). From the above description, the present invention does not need to test each macroblock and all prediction modes to find the minimum cost prediction. The mode is mainly divided into two steps, each step is only tested with several prediction modes of small-types, so the prediction time is significantly shortened, and the entire compression time can be greatly improved. Brief Description: The first figure is the compression flow chart of the invention applied to the H.264/AVC compressed image standard. The second figure is the detailed flow chart of the first figure. The third figure A 'B: the system is respectively hair The previous one used for compression and the giant block currently used. 1310654 The fourth picture is a block diagram of the compression standard. The fifth picture is the compression process of the standard .264/AVC compressed image standard. [Main component symbol description] (3 0 ) prediction module (3 1 ) external prediction unit (3 2 ) internal prediction early element (3 3 ) mode decision early element (40) conversion coding unit (4 1 ) converter
(4 2 )量化器 (4 4 )減法器 (5 1 )反轉換器 (5 3 )加法器 (4 3 )嫡編碼器 (5 0 )還原單元 (5 2 )反量化器 (5 4 )解除方塊效應濾波器 15(4 2 ) quantizer (4 4 ) subtractor (5 1 ) inverse converter (5 3 ) adder (4 3 ) 嫡 encoder (5 0 ) reduction unit (5 2 ) inverse quantizer (5 4 ) release Square effect filter 15