201117618 六、發明說明: 【發明所屬之技術領域】 本發明是有關於一種可調性視訊編碼方法,且特別是有關於 一種依據可調性設定所對應之使用者分配率,進行各次頻帶之資 料率分配的方法。 【先前技術】 多重傳播(multicast)技術可將多媒體資料經由網際網路同時 〇 地傳送至眾多使用者。然而,在有限的頻寬下,多重傳播技術面 臨了因應不同顯示解析度、播放率以及顯示品質而提供視訊内容 的挑戰。近幾年來’可調性視訊編解碼器(scalablevide〇c〇dec) 發展促使多媒體視訊可多重傳播於非齊性網路環境之中。可調性 視訊編解碼器除了可採用小波(wavelet)表示方式中所具之多重 解析度特性而獲得空域(spatial)及時域(temporal)上的可調性 (scalability )’也可採用基於三維小波轉換之動態補償時域濾波處 理來獲得空域、時域及訊雜比上的可調性。 在三維小波編碼設計中’將視訊序列空域及時域上進行小波轉 〇 換處理可分解產生多個空時域次頻帶,而編碼策略則可視為在這 些空時域次頻帶(Spatial-temporal subband )進行位元分配的問題。 在現今常用的三維來源編碼器之中,對空時域次頻帶位元分配進 行最佳化的效果並不顯著,這一類的來源編碼器為基於應用於空 時域小波轉換的小波濾、波器為正交(orthogonal)的條件,假設經 空時域小波轉換前後會能量守恆來進行處理。 然而,當採用雙正交(bi-orthogonal)小波濾波器及/或時域濾 波處理所需之動態補償方式時,能量守恆的假設便不適宜,其中 雙正交小波濾波器,例如:5-3濾波器或9_7濾波器,其僅能維持 201117618 經空時域小轉猶魏_能量技,__ 成像素值域及小錄域上能量⑽變。耻,考量 轉換所造成的能量改變,便有人提出於各次頻帶量化誤 二 不同權重的分析來解決能量改變問題,又單 : 遞模型(_卿agatkmm()del)延伸至傳 ❹ Ο 列。5 3送端施依據<閱使用者的資訊來編瑪視訊序 【發明内容】 對#可雛舰編财法,其可依射雛設定所 對應之使財分配率’進行次鮮之f料率分配。 人之ϋ提$種可雛視訊編碼方法。首先,將視訊序列所包 行二維小波轉換處理及動態補償時域濾、波處理, 慮理接帶。依據二維小波轉換處理及動態補償時域渡波 相依錯誤傳遞模型,並據以分析獲得各次頻帶之 =、緣。而且,依據視訊編解碼器支援之多個可調性設 ^ 上述次鮮巾取得可嫩設定分別所需之多個次頻帶集 雛設定分別具有多個使用者分配率。在預定頻寬下’ 率及各次娜所對應之糾域權重絲,提供一 最佳化處理,便可獲得各次頻帶 5 201117618 > ^ 在本發明之-實施射可雛設 “a: #_•度' S種播放率及多種顯示品質之排列組合。 上述之可雛視訊編碼方法’在本發狀-實酬巾次頻帶之 1域權f m相關於二維小波轉換處理所顧之小波係數以及 動態麵時域濾波處理所採用之移動向量。 上述之可雛視訊編碼方法’在本發明之_實施射提供目標 函之步驟包括在各次頻帶集合下,計算各次頻帶之一第一統計 值與其所對應之空時域權重係數相乘積的第一權重和 ,且在這些 〇可^性奴下’計算各次頻帶集合所對應之第-權重和與各可調 性叙所對應之使用者分配率相乘積的第二權重和,其中目標函 f k視訊序列之第二崎值與第二權重和賴應關係。 〜基於上述’本發明實齡丨考量則、_麟理及喊紐處理 别後能量改變的問題’提供轉相依錯誤傳觀型並且據以分析 獲得次頻帶之空時域權重係數。此^時域權重魏相關於二維小 波轉換處理所_之小波絲及動態麵時賴祕理所採用之 移動向量’藉以更精確地估測次小波值域上的失真程度。另外, 為了目應不㈤的視訊需求,更考量訂閱各種可雜設定之使用者 Ό 分配率來進行次頻帶之資料率分配。 【實施方式】 圖1繪示為本發明之一實施例之可調性視訊編碼方法的方塊 圖。請參照圖1,首先’伺服端110將視訊序列120所包含之多個 里面進行一維小波轉換處理(t^Yo-dimension waveiet壮咖免血), 藉以將各畫面分解為多個次頻帶厂石,其中r表示視訊序列i2〇之 中的晝面索引,且/^表示畫面/rr於第k層空域分解中的第j個次 頻帶。接著,本實施例基於以偶畫面預測奇畫面實現之動態補償, 201117618 將各次頻帶/¾進行動態補償時域遽波處理(m〇ti〇n c〇mpensated temporal filtering,MCTF),以產生多個高通次頻帶<+ι及低通二欠201117618 VI. Description of the Invention: [Technical Field] The present invention relates to a tunable video coding method, and more particularly to a frequency band corresponding to a user allocation rate according to an adjustability setting. Method of data rate allocation. [Prior Art] Multicast technology can simultaneously transmit multimedia data to many users via the Internet. However, with limited bandwidth, multi-cast technology faces the challenge of providing video content in response to different display resolutions, playback rates, and display quality. In recent years, the development of scalable video codecs (scalablevide〇c〇dec) has enabled multimedia video to be multi-distributed in non-homogeneous network environments. The tunable video codec can obtain spatial scalability on the temporal basis in addition to the multiple resolution characteristics of the wavelet representation. It can also be based on three-dimensional wavelets. The dynamic compensated time domain filtering process of the conversion obtains the adjustability in the airspace, time domain, and signal-to-noise ratio. In the 3D wavelet coding design, the wavelet transform-transformation process in the spatial domain of the video sequence can be decomposed to generate multiple spatio-temporal sub-bands, and the coding strategy can be regarded as the spatial-temporal subband. The problem of bit allocation. Among the commonly used three-dimensional source encoders, the optimization of the space-time sub-band bit allocation is not significant. This type of source encoder is based on wavelet filtering and wave applied to space-time domain wavelet transform. The device is an orthogonal condition, and it is assumed that the energy conservation is performed before and after the wavelet transform in the space-time domain. However, when the dynamic compensation method required for bi-orthogonal wavelet filter and/or time domain filtering is used, the assumption of energy conservation is not suitable, and the bi-orthogonal wavelet filter, for example: 5- 3 filter or 9_7 filter, which can only maintain the 201117618 space-time domain small turn wei _ energy technology, __ into the pixel range and the small recording field energy (10) change. Shame, considering the energy change caused by the conversion, it has been proposed to solve the energy change problem by analyzing the different weights of each frequency band, and the single-handed model (_qing agatkmm() del) extends to the Ο Ο column. 5 3Send the end of the application according to the user's information to compile the video order [invention content] ## 雏 舰 编 编 编 编 对 雏 雏 雏 雏 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰 舰Rate allocation. The person can raise the video coding method. First, the video sequence is packaged by two-dimensional wavelet transform processing and dynamic compensation time domain filtering and wave processing. According to the two-dimensional wavelet transform processing and the dynamic compensation time-domain wave-dependent error-conveying model, and according to the analysis, the sub-bands and edges are obtained. Moreover, according to the plurality of adjustability settings supported by the video codec, the plurality of sub-band set settings required for obtaining the sensible setting respectively have a plurality of user allocation rates. In the predetermined bandwidth, the rate and the correction domain weights corresponding to each sub-N, provide an optimization process to obtain the sub-bands 5 201117618 > ^ In the present invention - the implementation of the shootable "a: #_•度' S-play rate and a combination of multiple display qualities. The above-mentioned video encoding method '1 in the sub-band of the hair-real-receiving towel sub-band fm related to the two-dimensional wavelet transform processing The wavelet coefficients and the motion vectors used in the dynamic surface time domain filtering process. The above-described video encoding method of the present invention includes the step of providing a target function in the present invention, including calculating one of each frequency band under each frequency band set. a first weight sum of a statistic value multiplied by its corresponding null time domain weight coefficient, and under these succumbs, the first-weight sum and the tunable refinement corresponding to each sub-band set are calculated Corresponding user weight distribution product of the second weight sum, wherein the second fetch value of the target function fk video sequence is related to the second weight and the dependence. ~ Based on the above-mentioned 'the actual age of the present invention, _ And shouting to deal with energy changes The problem 'provides a phase-dependent error propagation type and analyzes the null time domain weight coefficient of the sub-band. This ^ time domain weight is related to the wavelet and dynamic surface time of the two-dimensional wavelet transform processing. The motion vector' is used to estimate the degree of distortion on the sub-wavelength range more accurately. In addition, in order to meet the video requirements of (5), it is necessary to consider the user's allocation rate of various miscellaneous settings to calculate the data rate of the sub-band. [Embodiment] FIG. 1 is a block diagram of a tunable video encoding method according to an embodiment of the present invention. Referring to FIG. 1, first, the server 110 performs a plurality of contents included in the video sequence 120. Dimensional wavelet transform processing (t^Yo-dimension waveiet), which is used to decompose each picture into multiple sub-band factories, where r represents the facet index in the video sequence i2〇, and /^ indicates the picture/ Rr is the jth sub-band in the k-th layer spatial decomposition. Next, the present embodiment is based on the dynamic compensation realized by the even picture prediction odd picture, and 201117618 performs dynamic compensation time-domain chopping processing for each frequency band/3⁄4. m〇ti〇n c〇mpensated temporal filtering, MCTF), to produce a plurality of high-pass subband < + ι and low pass under two
頻帶l^·,其中動態補償時域濾波處理可採用5_3濾波器或者9 7 濾波器實現之。 B 圖2、緣示為本發明之一實施例之動態補償時域遽波處理的示 意圖、。請參闕2,_用5_3濾波器實現之動_償時域遽波處 理且進行兩層時域小波分解為例,高通次頻帶增+1以下列等式⑴ 表示之: 其中,V及V分別為全解析度下的向前移動向量及向後移動向 量,且ν々· = ν/2々為第々·個次頻帶的移動向量。等 續及f的内插方式並未考慮交叉相位之小波係帶 於此,本實施例採用a tr〇us (AT)演算法所提供之最佳動態 補償濾、波架構,因此高通次頻帶碎+i及低通次頻帶1^可分別以 下列等式(2)及(3)表示之: 〇 H-+l(P) = ^\p)-~(AT_F^p + v^ + AT_F2i^ 4+100=增_⑶ 其中,义7^為次頻帶乂,交錯過完備小波係數(如池福 〇v_mplete wavelet coefflcient)’jr^(2i + v)為於(2^〇)位 置上的内插值,且^^為為高通次頻帶七的過完備小波係數。 從等式(1)及(2)可以觀察到像素值域的能量經空域小波轉換、時 域小波轉換以及動態補償時域濾波的移動估測處理後會有所改 變,以下說明本實施例為維持像素值域與小波值域之間的能量守 恆,逐-地推衍獲得空域權重係數、時域權重係數以及空時域權 201117618 重係數。 在二維小波轉換處理中,基於量化誤差為白雜訊且其與系統隨 機雜訊互不相關的假設,經重建之各晝面於像素值域下的統計值 (例如:均方誤差值)為各次頻帶於小波值域下統計值(例如: 均方誤差值)的權重和,其中各次頻帶之空域權重係數相依於小 波係數。以多層空域小波分解為例,經重建之畫面的均方誤差值 如下列等式(4)所示: σ/=Σ^·σ|--(4) Ο 其中1^々為各次頻帶所對應之空域權重係數,且σ|·為各次頻帶之 均方誤差值。於此,空域權重係數5^^·取決於小波轉換處理所採 用之濾波器,當採用正交濾波器時,空域權重係數5你尽便為i。 在動態補償時域遽波處理中,本實施例以矩陣p表示預測 (prediction)步驟所採用之轉換矩陣,且以矩陣u表示更新 (update)步驟所採用之轉換矩陣,藉以產生高通次頻帶月#+1及 低通次頻帶1¾。請參照圖2,採用單層動態補償時域濾波處理的 條件下,等式(2)及(3)可分別以等式(5)及(6)重新表示之: Λ2,+1 = /2ί+1 — (尸 2ί,+/2ί + 尸2/+2/2/+2 ) _ _(5) 〇 l2i =f2i+(U2i^h2i~l +U2i^+h2i+l)-(6) 其中,A及/分別為咼通次頻帶之行向量(column vector)及 低通次頻帶之行向#,P2卜及之上標分職示基於第錄畫 面的移動方向,且向前預測及向後預測分別標示為+及―。於此, 可觀察到時域濾波處理不僅與所選擇之小波濾波器係數相依,亦 會受到動態資訊影響’例如:轉換矩陣ρ及U。 有鑑於此,本實施例依據上述之二維小波轉換處理及動態補 償時域濾波處理,提供了動態相依錯誤傳遞模型。在此動態相依 錯誤傳遞模型中,經重建之奇晝面的量化誤差4产及偶畫^的量 201117618 化誤差可分別以下列等式(7)及(8)表示之: Δ/2ι· =Δ/2ι· ~U2i^Ah2M ..(7) △严1 = ΔΑ2ί+1 -尸21,+△产一尸2/+2,-△ 一⑻ = 及”別為低通次頻帶及高通次頻帶經失真來源編碼所 產生之ϊ化誤差。 Ο 接著’依據動態相依錯誤傳遞模型,更進一步地推衍經重建 面的均方誤差值,於此可觀察到某—次頻帶上的單位誤差 ^傳遞至像素值域上多張畫面之誤差。以單層_補償時域處理 為例’,重建之各畫面的均方誤差值如下列等式⑼所示,即: σ -2/-2 2r, /2卜1 σ/2ί _2 σ/2ί+, σ2 σ严2 ^ y2/+3 2 σ/2Μ σ 1-^0 cr h 2/-3 ;2/-2 σ 2 h σ; 2/-1 /2/ 2ί+1 2 σ -(9) /2/+2 h 2/+3 Ο 其中為權重矩陣,且權重矩陣yl—O中的元素〇〇及 分別表示/’受到次頻帶y及次頻帶〆單位誤差的權重影 響程度。於此,經重建之各畫面於像素值域下的統計值(例如: 均方誤差值)為各尚通次頻帶及各低通次頻帶於小波 值(例如:均方誤差值)的權重和,且時域權重係數彻可以下 列等式(10)表示之: -(10) 201117618 X<r° ΣΧ?0 ςΧγ0 = Σ〆:0 twl~^°(h2i+l) twl^°(l2i+2) • • Ο 同理,以多層動態補償時域濾波處理為例,第1層時域分解所 重建之訊號的均方誤差值及時域權重係數分別如下列等式(11)及 (12)所示: Ο h2㈠ 2 /2ί-2 uh2i~l _2 ah2M _2 〇 j2i+2 O' 1,2/+3 w 2— aih2i~6 2 Ίι2 •2 7Λ' CT1/2/-4 O' 4« O' II.2/+2 -(11) .2 lhu σ//2ί+4 .2 lh2 O' 11,2/+3 tw2^°(lh2i~6) tw2~^0(ll2i-4) tw2~^°(lh2i~2) tw2^°(lh2i+2) tw2~^°(l2i+4) TqwqY0 Σ,^° = Σ,^° Σ,々0 • • * -(12) #由上述等式(11)及(12)可以得知,在第個時域次頻帶中第衫 個空域次_的空時域權重雜可從线權重絲及時域權重係 201117618 ·· 數相乘而獲得之,亦即空時域權重係數 伽〆0 (" 2’) = *5、X 伙2—0 (" 2ί·)。 依據資料率失真(rate-distortion,R-D)理論,須先對造成重 建之畫面最大失真的次頻帶進行位元分配。本實施例利用上述推 竹所付之空時域權重係數,得以在小波值域上更精確地估算失真 程度,進而獲得各次頻帶較佳的資料率分配。為了因應使用者不 同的視訊需求(例如:解析度、播放率以及顯示品質),可調性視 訊編解碼器通常會提供多種可調性設定,而這些可調性設定分別 0 有其對應之使用者分配率,其中這些可調性設定例如分別為上述 多種解析度、播放率以及顯示品質之組合。於此,本實施例依據 各可調性設定,從上述次頻帶中取得各可調性設定所需之次頻帶 集合。 圖3繪示為本發明之一實施例之不同可調性設定所對應之使 用者分配率的示意圖。請參照圖3A,某一畫面經2層空域小波轉 換而分解成7個次頻帶SP0〜SP6。透過次頻帶SP0〜SP6之重建, 可提供1/4QCIF、QCIF以及CIF三種解析度之影像至客戶端,且 這些可調性設定分別有其對應之使用者分配率穴1)、穴2)及穴3)。 〇 也就是說,30%的使用者訂閱1/4QCIF之影像,70%的使用者訂閱 QCIF之影像’ 0%的使用者訂閱CIF之影像。1/4QCIF之可調性設 疋所需之次頻帶集合Swqqf包含次頻帶SP0,QCIF之可調性設定 所需之次頻帶集合S〇cIF包含次頻帶SP0〜SP2,且CIF之可調性設 定所需之次頻帶集合SCIF包含次頻帶SP0〜SP6。 請參照圖1’為了充分地利用可用的預定頻寬以及提供使用者 多種可適性設定之影像,本實施例依據使用者分配率^灸)'及各次 頻帶所對應之空時域權重係數5^^而提供一目標函數,經由將此目 標函數進行最佳化處理130 ’例如:動態規劃最佳化處理,來得到 201117618 解’並且經由網路i5G多重傳射料流140 者分触說,在預賴寬τ,本實補依據使用 率洲产各次頻帶所對應之空時域權重係數提供目標 函數/ = Γηίη^_··Γ/) }來 财基於資辭&之咐 設定下所需蝴數’物第_可調性 Ο 域’本領域Μ此㈣者也可_依料行祕補償時 域遽波處理及二維小波轉換處理之議架構來產生次頻帶。 檢-ΐΐί述實施例所述,於此可歸納為下列的方法絲。圖4 昭—實施例之可雛視訊編碼方法的流程圖。請參 j 1及圖4 ’首先,提供視訊序列12〇 (步驟s4〇i ),其包含多 请、接著將各晝面進行二維小波轉換處理及動態補償時域 以產生多個次頻帶(步驟S4G2)。在預定頻寬下 ,依據 所對應之使用者分配率以及各次頻帶之空時域權重係 ^ =供目標函數(麵S403 ),其中依據可調性設定,可從次頻 可調性設定所需之次頻帶集合,且空時域權重係數為經 依錯誤傳遞翻所獲得之。經由將此目標函數進行最 佳化處理’便可獲得各次頻帶應分配之資料率(步驟S404)。 所述,上述實施例考量經小波轉換處理及時域遽波處理 改變_題’提供動態她錯誤侧模型,並且據以分 j付次頻帶之空時域權重係數。利用此空時域權重係數,可 ^估斗欠小波值域上的失真程度。另外,為了因應不同的 寻》fU、、’上述實施例更考量訂閱各種可適性奴之·者分配 12 201117618 率來進行次頻帶之資料率分配,以充分地_可用類寬。 【圖式簡單說明】 圖1緣示為本發明之—實施例之可調性視訊編碼方法的方塊 圖2緣示為本發明之—實施例之動態補償時域濾波處理的示 Ο 圖3緣示為本發明之一實施例之不同可調性 用者分配率的示意圖。 設定所對應之使 圖4綠示為本個之—實酬之可雛魏編封法的流程 【主要元件符號說明】 110 :伺服端 120 .視訊序列 130 :最佳化處理 140 :資料流 150 :網路The frequency band l^·, wherein the dynamic compensation time domain filtering process can be implemented by a 5_3 filter or a 9 7 filter. B. Fig. 2 is a schematic diagram showing the dynamic compensation time domain chopping process according to an embodiment of the present invention. Please refer to , 2, _ using 5_3 filter to achieve dynamic _ time domain chopping processing and two-layer time domain wavelet decomposition as an example, Qualcomm subband increase +1 is expressed by the following equation (1): where, V and V They are the forward motion vector and the backward motion vector at full resolution, respectively, and ν 々 · = ν/2 々 is the motion vector of the second sub-band. The interpolation method of the continuation and f does not consider the wavelet phase of the cross phase. In this embodiment, the best dynamic compensation filter and wave architecture provided by the a tr〇us (AT) algorithm is used, so the Qualcomm sub-band is broken. The +i and low-pass sub-bands 1^ can be expressed by the following equations (2) and (3), respectively: 〇H-+l(P) = ^\p)-~(AT_F^p + v^ + AT_F2i^ 4+100=增增_(3) where, meaning 7^ is the sub-band 乂, interlaced over-complete wavelet coefficients (such as Chifu〇v_mplete wavelet coefflcient) 'jr^(2i + v) is at (2^〇) position Interpolation, and ^^ is the overcomplete wavelet coefficient of the high-pass sub-band 7. From the equations (1) and (2), the energy of the pixel range can be observed through spatial wavelet transform, time domain wavelet transform and dynamic compensation time domain filtering. After the motion estimation process is changed, the following description shows that the energy conservation between the pixel value range and the wavelet value domain is maintained, and the spatial weight coefficient, the time domain weight coefficient, and the space time domain weight are obtained by the derivation. 201117618 Heavy coefficient. In the two-dimensional wavelet transform processing, based on the assumption that the quantization error is white noise and it is not related to the system random noise, The statistical values of the reconstructions under the pixel value domain (for example, the mean square error value) are the weights of the statistical values of the sub-bands in the wavelet domain (for example, the mean square error value), and the airspace of each sub-band The weight coefficient depends on the wavelet coefficient. Taking the multi-space spatial wavelet decomposition as an example, the mean square error of the reconstructed picture is as shown in the following equation (4): σ/=Σ··σ|--(4) Ο 1 ^々 is the spatial weight coefficient corresponding to each frequency band, and σ|· is the mean square error value of each frequency band. Here, the spatial weight coefficient 5^^· depends on the filter used in the wavelet transform processing, when In the case of the orthogonal filter, the spatial weight coefficient 5 is as good as i. In the dynamic compensation time domain chopping process, this embodiment uses the matrix p to represent the transformation matrix used in the prediction step, and the matrix u represents the update. (update) the conversion matrix used in the step to generate the high-pass sub-band month #+1 and the low-pass sub-band 13⁄4. Please refer to Figure 2, using a single-layer dynamic compensation time-domain filtering process, equation (2) and (3) Re-represented by equations (5) and (6): Λ2, +1 = /2ί+1 — (尸2ί,+/2ί + 尸2/+2/2/+2) _ _(5) 〇l2i =f2i+(U2i^h2i~l +U2i^+h2i+l)-(6) where A And / respectively, the row vector of the sub-band and the line of the low-pass sub-band are #, P2 and above, based on the moving direction of the recorded picture, and the forward prediction and the backward prediction are respectively Marked as + and ―. Here, it can be observed that the time domain filtering process is not only dependent on the selected wavelet filter coefficients, but also affected by dynamic information 'for example: conversion matrices ρ and U. In view of this, the present embodiment provides a dynamic dependent error transfer model based on the above two-dimensional wavelet transform processing and dynamic compensation time domain filtering processing. In this dynamic dependent error transfer model, the quantized error 4 of the reconstructed odd plane and the amount of the even graph 201117618 can be expressed by the following equations (7) and (8), respectively: Δ/2ι· = Δ/2ι· ~U2i^Ah2M ..(7) △ strict 1 = ΔΑ2ί+1 - corpse 21, + △ producing a corpse 2/+2, -△ one (8) = and "not a low pass sub-band and high pass The demodulation error generated by the frequency band is encoded by the distortion source. Ο Then, according to the dynamic dependent error transfer model, the mean square error value of the reconstructed surface is further derived, and the unit error on a certain sub-band can be observed. The error is transmitted to multiple pictures on the pixel value domain. Taking the single-layer _compensation time domain processing as an example, the mean square error value of each reconstructed picture is as shown in the following equation (9), namely: σ -2/-2 2r , /2b1 σ/2ί _2 σ/2ί+, σ2 σ strict 2 ^ y2/+3 2 σ/2Μ σ 1-^0 cr h 2/-3 ; 2/-2 σ 2 h σ; 2/ -1 /2/ 2ί+1 2 σ -(9) /2/+2 h 2/+3 Ο where is the weight matrix, and the elements 〇〇 in the weight matrix yl-O and respectively represent /' subject to the sub-band y And the degree of weighting of the sub-band 〆 unit error. Here, the reconstructed paintings The statistical value in the pixel value domain (for example, the mean square error value) is the weight of each wavelet sub-band and each low-pass sub-band in the wavelet value (for example, the mean square error value), and the time domain weight coefficient can be The following equation (10) indicates: -(10) 201117618 X<r° ΣΧ?0 ςΧγ0 = Σ〆:0 twl~^°(h2i+l) twl^°(l2i+2) • • Ο Similarly, Taking the multi-layer dynamic compensation time domain filtering process as an example, the mean square error value and the time domain weighting coefficient of the signal reconstructed by the first layer time domain decomposition are respectively shown in the following equations (11) and (12): Ο h2(1) 2 /2ί -2 uh2i~l _2 ah2M _2 〇j2i+2 O' 1,2/+3 w 2— aih2i~6 2 Ίι2 •2 7Λ' CT1/2/-4 O' 4« O' II.2/+2 -(11) .2 lhu σ//2ί+4 .2 lh2 O' 11,2/+3 tw2^°(lh2i~6) tw2~^0(ll2i-4) tw2~^°(lh2i~2) Tw2^°(lh2i+2) tw2~^°(l2i+4) TqwqY0 Σ,^° = Σ,^° Σ,々0 • • * -(12) #from the above equations (11) and (12) It can be known that in the first time domain sub-band, the space-time weights of the first-order airspace sub-times can be obtained by multiplying the line weights and the time domain weights 201117618 ··, that is, the space time domain weights 〆0 coefficient gamma (" 2 ') = * 5, X partner 2-0 (" 2ί ·). According to the rate-distortion (R-D) theory, bit allocation must be performed on the sub-band that causes the maximum distortion of the reconstructed picture. In this embodiment, by using the space-time weighting coefficients paid by the above-mentioned push bamboo, the degree of distortion can be more accurately estimated in the wavelet domain, and the better data rate allocation of each sub-band is obtained. In order to respond to different video requirements of users (such as resolution, playback rate and display quality), adjustable video codecs usually provide a variety of adjustable settings, and these adjustable settings have their corresponding use of 0 The allocation ratio, wherein the adjustability settings are, for example, a combination of the above various resolutions, play rates, and display qualities. Herein, in this embodiment, the sub-band set required for each adjustability setting is obtained from the sub-band according to each adjustability setting. 3 is a schematic diagram showing a user allocation ratio corresponding to different adjustability settings according to an embodiment of the present invention. Referring to Fig. 3A, a certain picture is decomposed into seven sub-bands SP0 to SP6 by two-layer spatial wavelet transform. Through the reconstruction of the sub-band SP0~SP6, images of 1/4QCIF, QCIF and CIF can be provided to the client, and these adjustable settings respectively have their corresponding user allocation rate points 1), 2) and Hole 3). 〇 In other words, 30% of users subscribe to 1/4QCIF images, and 70% of users subscribe to QCIF images. 0% of users subscribe to CIF images. 1/4QCIF adjustability setting The required sub-band set Swqqf includes sub-band SP0, QCIF adjustability setting sub-band set S〇cIF includes sub-band SP0~SP2, and CIF adjustability setting The required subband set SCIF includes subbands SP0~SP6. Please refer to FIG. 1 ' in order to make full use of the available predetermined bandwidth and provide users with various adaptability settings, the embodiment according to the user allocation rate ^ moxibustion) 'and the frequency and time domain weight coefficient corresponding to each frequency band 5 ^^ provides an objective function, by optimizing the objective function 130 'for example: dynamic programming optimization processing, to obtain 201117618 solution ' and through the network i5G multiple transmission stream 140, said In the pre-requisite width τ, this real complement provides the objective function based on the space-time weighting coefficient corresponding to each frequency band of the continent. / = Γ ί ί ί ί ί ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) The required number of objects 'objects _ tunable Ο domain' in this field (4) can also be used to generate the sub-band according to the structure of the time domain chopping process and the two-dimensional wavelet transform process. This is described in the Examples, which can be summarized as the following method. Figure 4 is a flow chart of the video encoding method of the embodiment. Please refer to j 1 and FIG. 4 ' Firstly, a video sequence 12 〇 (step s4 〇 i ) is provided, which includes multiple requests, then performs two-dimensional wavelet transform processing on each side and dynamically compensates the time domain to generate multiple sub-bands ( Step S4G2). At a predetermined bandwidth, according to the corresponding user allocation rate and the null time domain weight of each frequency band ^ = for the objective function (surface S403), wherein the secondary frequency adjustability setting can be set according to the adjustability setting The subband set is required, and the null time domain weight coefficient is obtained by error transfer. By optimizing the objective function, the data rate to be allocated for each sub-band can be obtained (step S404). In the above embodiment, the wavelet transform processing and the time domain chopping processing change _question provides a dynamic her error side model, and the null time domain weight coefficient of the subband is divided. Using this null time-domain weighting factor, the degree of distortion on the under-wavelength range can be estimated. In addition, in order to respond to different homing fU,, and the above embodiments, the data rate allocation of the sub-band is performed to more fully subscribe to the data rate distribution of the sub-band. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an adjustable video encoding method according to an embodiment of the present invention. FIG. 2 is a schematic diagram of a dynamic compensation time domain filtering process according to an embodiment of the present invention. Shown is a schematic diagram of different adjustable user allocation rates for an embodiment of the present invention. The corresponding setting makes the green of Figure 4 as the one--the flow of the real-life can be edited and sealed [main symbol description] 110: server 120. video sequence 130: optimization processing 140: data stream 150 :network
161〜162:客戶端 %、轉+1、乓:次頻帶 P、U:轉換矩陣 K幻:使用者分配率 SP0〜SP6 :次頻帶 0广〇6 :使用者分配率161~162: Client %, turn +1, pong: sub-band P, U: conversion matrix K-magic: user allocation rate SP0~SP6: sub-band 0 〇6: user allocation rate
Sl/4QCIF、SqCIF、sCIF :二欠頻帶集合 7〜本發狀-實酬q概視職财法的各步 13Sl/4QCIF, SqCIF, sCIF: Set of two underbands 7~This issue--Remuneration q Overview of the steps of the business method 13