TWI232684B - Apparatus for variable bit rate control in video compression and target bit allocator thereof - Google Patents
Apparatus for variable bit rate control in video compression and target bit allocator thereof Download PDFInfo
- Publication number
- TWI232684B TWI232684B TW92120340A TW92120340A TWI232684B TW I232684 B TWI232684 B TW I232684B TW 92120340 A TW92120340 A TW 92120340A TW 92120340 A TW92120340 A TW 92120340A TW I232684 B TWI232684 B TW I232684B
- Authority
- TW
- Taiwan
- Prior art keywords
- complexity
- day
- mentioned
- current
- instantaneous
- Prior art date
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
Description
12326841232684
【發明所屬之技術領域】 特別係指一種適用 之裝置。 本發明係有關於資料壓縮之領域, 於視訊編碼系統以執行可變位元率控制 【先前技術】 眾所周知,電子通訊世界正經歷一次數位革命,以數 位表示資訊的主要優點在於能夠幾近無誤地儲存、再生、 傳收、處理及運用的資料位元流(bitstream)。舉例而言 ,NTSC的彩色視訊影像每秒有29· 97張晝面,每張書面^[Technical field to which the invention belongs] Particularly refers to a suitable device. The present invention relates to the field of data compression. It performs variable bit rate control in video coding systems. [Prior art] As we all know, the world of electronic communication is undergoing a digital revolution. The main advantage of representing information digitally is that it can be almost error-free. A bitstream of data that is stored, reproduced, transmitted, processed, and used. For example, NTSC's color video image has 29.97 daylights per second, each written ^
4 8 0條可見掃瞄線’每條掃瞄線約需4 8 〇點以紅、綠、誃 '三 色呈現的像素(pixel),但若每種色彩成分以8位元編;了 則所產生的位元率(bitrate)每秒約168百萬位元(Mbits/ 5 ),故各種視訊格式其未經壓縮處理的位元率是非常高而 不經濟,因此不適於多數的應用。 與電腦、電信網路、消費性產品整合的數位音訊和視 5孔,更加刺激資成革命的前進,而這革命的核心則是視、 音祝的數位壓縮技術。許多的壓縮標準,包含以壓縮技術 共通之精髓為基礎的演算法,如:I TU-T(前身係CC ITT)建 議書 H.261 和 EL 2 63,以及 ISO/IEC 的 MPEG-1、MPEG-2 和480 visible scan lines' Each scan line requires approximately 480 points of pixels in three colors: red, green, and 誃, but if each color component is edited in 8 bits, then The generated bit rate is about 168 million bits per second (Mbits / 5). Therefore, the uncompressed bit rate of various video formats is very high and uneconomical, so it is not suitable for most applications. The digital audio and video channels integrated with computers, telecommunications networks, and consumer products have further stimulated the advancement of the revolution, and the core of this revolution is digital compression technology for video and audio. Many compression standards, including algorithms based on the essence of common compression technology, such as: I TU-T (formerly CC ITT) Recommendations H.261 and EL 2 63, and ISO / IEC MPEG-1, MPEG -2 and
MPEG-4標準。MPEG的演算法是由動態影像專家群組 (Moving Picture Experts Group,MPEG)所發展出來,該 動態影像專家群組係國際標準組織(I n t e r n a t i ο n a 1MPEG-4 standard. The MPEG algorithm was developed by the Moving Picture Experts Group (MPEG), which is an international standard organization (I n t e r n a t i ο n a 1
Standards Organization,ISO)及國際電子技術委員會 (International Electrotechnical Commission,IEC)的Standards Organization (ISO) and International Electrotechnical Commission (IEC)
m mm m
0702-7606twF(nl) ; 90P132 ; Lm.ptd 第6頁 12326840702-7606twF (nl); 90P132; Lm.ptd Page 6 1232684
二二技術委員會,致力於發展視、音訊的壓縮及多路傳殊 、、表^方式,這些標準規定了壓縮位元流的語法(syntax) 兔及解碼的方法’但對於編碼器所使用的演算法而言,卻 為編碼的新穎性與多樣性保留了相當多的自由度。 、查根據,一連串的視訊畫面(pi cture)分成一序列 =旦面群組(group of picture,G〇p),其中每組G〇p以卜 ,面開始,後面跟著p—畫面和β—晝面的安排,第工圖以顯 二的順序5兒明一組典型的G〇p。卜晝面之編碼毋須參考先 前,將來的晝面,P-晝面則參考連串的視訊畫面中在時間 上最接近的I-晝面或卜晝面來進行編碼,而B_畫面散佈於 I -畫面和P-晝面之間。並且利用先前的、將來的或兩者兼 具的^鄰I-晝面和P-畫面來編碼。雖然好幾張β—晝面可以 緊接,連續出現,但絕不能以B—晝面預估其他的晝面。 一每張晝面具有三種成分:亮度值(luminance,以γ表 示)’紅色差值(red color difference,以cr表示),以 及藍色差值(blue col〇r difference,以⑶表示)。對於 MPEG的4 : 2 : 0的格式而言,每一種Cr和⑶成分的取樣點在 水平與垂直方向只有γ成分的一半。如第2圖所*,一張 MPEG的畫面其基本構成方塊係大區塊i〇ck,以〇 表示)。以4:2:0的視訊為例,每個〇包含一個¥成分16}[16 取樣點的陣列,以及兩個Cr和Cb成分8x8取樣點的區塊, 其中Y成分16x16取樣點的陣列實際上由4個8χ8取樣點的區 塊所組成。 、、扁馬的作用在於決定何種晝面編碼型態以及何種預The 22 technical committee is dedicated to the development of video and audio compression and multi-channel transmission. These standards specify the syntax of the compressed bit stream and the method of decoding. But for the encoder used As far as the algorithm is concerned, it retains a considerable amount of freedom for the novelty and diversity of coding. According to the investigation, a series of video pictures are divided into a sequence = group of picture (G0p), where each group of G0p starts with Bu, face, followed by p—picture and β— The arrangement of the day and the day, the first drawing shows a typical set of Gop in the order of two. The coding of the diurnal plane does not need to refer to the previous and future diurnal planes, and the P-diurnal plane refers to the I-dimensional plane or the diurnal plane that is closest in time in the series of video pictures to encode, and the B_ picture is scattered in Between I-picture and P-day. And use the previous, future or both ^ -neighbor I-day and P-picture to encode. Although several β-day surfaces can appear next to each other and appear continuously, the B-day surface must not be used to estimate other day surfaces. Each diurnal surface has three components: a brightness value (represented by γ), a red color difference (represented by cr), and a blue color difference (represented by ⑶). For the MPEG 4: 2: 0 format, the sampling points for each Cr and ⑶ component are only half of the γ component in the horizontal and vertical directions. As shown in Figure 2 *, an MPEG picture is basically composed of large blocks (ick, represented by 0). Taking 4: 2: 0 video as an example, each 0 contains an array of ¥ 16} [16 sampling points, and two blocks of 8x8 sampling points of Cr and Cb components, where the array of 16x16 sampling points of Y component is actually It consists of 4 blocks of 8 × 8 sampling points. The role of the flat horse is to determine what type of day-surface coding and what kind of prediction
1232684 五、發明說明(3) — ΐϊίίίί。對每張1_畫面,仙中的每個8x8方塊均經 放餘弦轉換(dlscrete c〇sine transf〇rn],κ ί 轉換係數陣列,轉換係數接著以一量化矩陣進行旦y 到二ί ΐ用z字形(zig_zag)掃猫Dc丁係數的量化結果而二 /一連串的DCT係數,並且此dCT係數序 . (如 able length c〇de,VLC)進 欠長度瑪 定將每個MB以I型MB或p型⑽進行 ='、須決 上述的方爷炎— 馬1型Μβ的編碼係以 前查面所ΐ 而對於每個Ρ麵,則需得到則Β以先 别!面所做之預測,此預測係藉由 β 乂先 vector)獲得,移動向量象徵著目前畫面動向里^〇t1〇n1232684 V. Description of Invention (3) — ΐϊίίίί. For each 1-picture, each 8x8 square in the fairy is subjected to cosine transform (dlscrete c〇sine transf rn), κ ί array of conversion coefficients, and the conversion coefficients are then subjected to a quantization matrix from y to two. zigzag (zig_zag) scans the quantized result of the Dc coefficients of the cat and two / a series of DCT coefficients, and this dCT coefficients are ordered. (such as the available length code, VLC) the length of each MB is set to I type MB Or p-type ==, depends on the above-mentioned Fang Yeyan — the coding of the horse type Mβ is previously checked, and for each P-plane, you need to get the prediction of B before you! This prediction is obtained by β (first vector), and the motion vector symbolizes the current picture movement ^ 〇t1〇n
及其在先前畫面中的預估Μβ之間的變_ 碼的MB 間的預測誤差則以DCT、量 預估MB與目前Μβ 碼。 Ζ予形掃瞄以及VLC進行編 處理Β-畫面時,必須決 刪:ί模式、F模式、β模式以_列何種MB模式來編碼每 身而不借助移動補償來編碼(如工 1式。I模式係以MB本 的向前預測編碼,係以先前書 一般);F模式為單向 型MB—般);反之,b模式為^ 侍到移動補償的預估(如P 來的晝面得到移動補償的預向後預測編碼,係以後 向的預測編碼,運用向前的和向$别的是FB模式,其係雙 做内插(interpolati〇n)而得° =的移動補償預估兩者來 對F、B和FB模式而言,預測誤果式之移動補償預估。 形掃猫以及VLC進行編碣。、 以使用DCT、量化、z字 視訊編碼器必須為整& t Μ & _ 里化位階以便在給定 0702-7606twF(nl) : 90P132 ; Lin.ptd 第8頁 1232684 五、發明說明(4) 制可見的失真,然而,以選取的量化位階編 會得知'里現實;數,:必然是在真正編碼之後才 達到的量化位階下,‘先、办f u數’能夠在給定希望 視訊編;;以:點2位元率的控制。簡中重要的 別是在固定位元率運:T :面所用位元數變化的限制’特 器檢驗者(nde=乍的f形,這種限制係透過視訊緩衝 每張書面而古VBVuLft 1 f 1 er,νβν)來施行。若對於 是以固定位^率^ 資料速率均相同,則視訊可說 虛擬緩衝器並且是Γ解;_編碼。νβν乃-種 位元率而言,編褐p分】二=;入緩衝器模型’以固定 缓衝器不會滿溢或置乏;:對位:額度須彻 元流則是以特定的吾士办對於了父位兀率,編碼過的位 、再沒有多的位元輸入為:率器直到其飽和 的位元率實際上是可錄 t 了解釋為進入VBV緩衝器 對於可變位元率,^ I "b達到上述的最大位元率。故 可變位元Ϊ控乏即可。 品質而還能維持輪出去糸I可能的提高解碼後視訊的 率的方式不同,可變位;圍内。和固定位元 位元數給景物較複雜、變動彈性來分配額外的 可變位元率控制架構卻相杏二…、的旦面。不過,傳統的 完成視訊編碼的處理,:雜,通常需進行好幾次才能 ,有鑑於此,亟需一種新穎的可變位 0702-7606twF(nl) ; 90P132 : Iin.ptd 第9頁 麵 1232684 五、發明說明(5) 元率控制技術,可用於„ ^ 亦期待能提供一種:人人…的硯訊編碼器,再者 .^ 禋通合施行於積體電路、農古丹者, 控制之動態影像壓縮枣 具有可變位元率 【發明内容】 根據本發明,每一張晝面其目標位 *品AA祕i ~的力配, 先前晝面的編碼結果和對ί將被編碼的目 愈 係基於 析的活動量,再者,藉由曰彳二而夕二二别量面其預先分 月恭乂奸庄去 猎由目則畫面之預先分析的活叙旦 及先别編碼畫面之實際複雜度間的關係,目動Ϊ以 度將能夠被估計而得。由於晝面的複雜度有^之複雜 之後的品質資訊,故此預估的目前畫面 面編碼 ^是非常細,可以更精確地分配目標位。J = 晝面,藉以達到視訊序列中不同晝面之間均一的二:張 質。 U見。口 本發明係針對於一種動態影像壓縮之可變位元率护 衣置。根據本發明,該裝置包括一移動估算單元、一複雜 =計器、-目標位元分派器以及一參數更新器。移動估 ::兀接收視訊序列裡2 -組連續畫面中即將被編碼的目 以計算此目!晝面之全活動量。基於目前晝面 ί = ! =及這組連續晝面中-同型態的先前編碼書面 面之複雜度。以目前畫面預器可用來估計目前晝 更新本組連續晝面之瞬間複雜户是雜度,目標位兀分派器 之既定平均位元率、統計複雜並且基於這組連續晝面 没及瞬間複雜度來計算該組The prediction error between the MB of the variable _ code and its estimated Mβ in the previous picture is DCT, the estimated MB and the current Mβ code. When Z-shape scanning and VLC editing and processing of B-pictures, it must be deleted: Which MB mode is used for ί mode, F mode, and β mode to encode each body without using motion compensation to encode (such as the type 1 method) .I mode is based on the MB forward prediction coding, which is based on the previous book); F mode is a one-way MB—like); on the other hand, mode b is ^ to estimate the motion compensation (such as the day of P) Pre-backward prediction coding with motion compensation in the front, which is backward prediction coding. The forward and backward $ FB modes are used, which are double interpolation (interpolati〇n) to get ° = motion compensation estimation. Both for F, B, and FB modes, the prediction error-like motion compensation estimation. Shape scan and VLC for editing. To use DCT, quantization, z video encoder must be integer & t Μ & _ Refine the ranks to give 0702-7606twF (nl): 90P132; Lin.ptd Page 8 1232684 V. Description of the invention (4) Make visible distortions, however, the selected quantization ranks will know 'In reality; number ,: inevitably at the quantization level reached after real encoding,' first, The fu number can be edited in a given desired video; with: point 2 bit rate control. The important thing in Jane is to operate at a fixed bit rate: T: the limit of the number of bits used by the plane. (nde = f shape at first glance, this restriction is implemented by video buffering each written and ancient VBVuLft 1 f 1 er, νβν). If the data rate is the same at a fixed bit ^ rate, the video can be said to be virtual The buffer is also a Γ solution; _ encoding. Νβν is the bit rate, and the brown p is divided into two]; into the buffer model 'to fix the buffer will not overflow or be depleted ;: Alignment: The quota must be completely streamed, and the specific bit rate is used for the parent bit rate. The encoded bits and no more bit inputs are: the bit rate is actually recordable until the bit rate is saturated. It is explained that for the variable bit rate into the VBV buffer, ^ I " b reaches the above-mentioned maximum bit rate. Therefore, the variable bit control is not sufficient. The quality can also be maintained, and I can improve decoding. The method of the rate of the rear video is different, variable position, in-field, and fixed-bit number. To allocate additional variable bit-rate control architectures, it is quite similar ... However, the traditional processing of video coding is usually complicated several times. In view of this, a new type of Variable position 0702-7606twF (nl); 90P132: Iin.ptd Page 9 1232684 V. Description of the invention (5) Element rate control technology, which can be used for ^ ^ We also look forward to providing a kind of encoder for everyone: , And more. ^ The Tongtong is applied to the integrated circuit and Nonggudan, and the controlled dynamic image compression has a variable bit rate. [Summary of the Invention] According to the present invention, the target position of each daytime surface is * AA. The power match of i ~, the previous coding results of the day and the target to be coded are based on the analysis of the amount of activity, and furthermore, by pretending to be sexually assaulted in advance by the second month and the second day. Zhuang went to hunt the relationship between the actual analysis of the pre-analyzed live picture and the actual complexity of the pre-coded picture, and the degree can be estimated by eye movement. Because the complexity of the day and the day has quality information after ^ is complex, the estimated current picture surface code ^ is very thin, and the target bits can be allocated more accurately. J = diurnal plane, in order to achieve a uniform two between different diurnal planes in the video sequence: Zhang Zhi. U see you. The present invention is directed to a variable bit rate protective garment for moving image compression. According to the present invention, the device includes a motion estimation unit, a complex counter, a -target bit dispatcher, and a parameter updater. Mobile Estimation :: The 2-group of consecutive pictures in the received video sequence to be coded to calculate this! Day-to-day activity. Based on the complexity of the current day surface ί =! = And the previously coded written facets of this set of consecutive day-surfaces of the same type. The current picture pre-processor can be used to estimate the momentary complexity of the current day-to-day update of the continuous day-to-day group of this group. The target average bit rate of the target dispatcher is complex and statistically complex. To calculate the group
0702-7606twF(nl) ; 90P132 ; Lin.ptd 五、發明說明(6) 連續畫面之瞬間位元率, 面=間複雜度及瞬::::晝面之複雜度、這組 目松位:額度分配給目前晝面。上目標位元分派器將 編碼之後’根據目前畫面之全二數更新器在目前畫面 編瑪目前畫面實際所用量化位階::、實際消耗位元數和 其活動量對複雜m·其中,值,計算目前畫面 度=比與目前畫面之全活動=晝面的活動量對複雜 際消耗位7L數以及目前晝面 ,而與目前畫面之實 反比二此外,參數更新器還以=寸=量化位階之平均值成 計算前述的統計複雜度。 則I面之複雜度為基礎來 另一方面,本發明還揭露一 一瞬間位元率計算器、一配額抑&位元分派器,包括 間位元率計算器接收視訊序列°中2 f、以及一限度單元。瞬 碼的目前晝面所含之複雜度以及該^連,畫面裡即將被編 度,=以該複雜度為基礎來更新:之統:複雜 度,並根據一函數來計算這組連 ^二畫面之瞬間複雜 中該函數係取決於該組連續之ς =瞬間位元率,其 複雜度及瞬間複雜度。配額元率、統計 ,位元額二 !率及瞬間複雜度為基礎,將目 :元額产f ::給目則畫面’而限度單元則用來將此目標 位几額度限制在既定的上限之下。 【實施方式】 為使本發明之上述目的、特徵和優點能更明顯易懂, (7> 五、發明說明 作詳細說明如 下文特舉一較佳實施例,並配合所附圖式 下: 抑一開始,預測用的卜晝面和P-晝面必須先在MPEG編碼 器内解碼,利用計算原始晝面及解碼的晝面間的均方根 (root mean square)誤差,可以得到解碼晝面其品質的客 觀量度。以均方根誤差作為解碼晝面之失真度,則解碼晝 面的複雜度(complex)-C定義如下: C = r X d 其中’ r係編碼該晝面所需的位元率,d係解碼的晝面之失 真度。理論上’平均的量化位階(q u a n t丨z a t丨〇 n s t e ρ s i z e)和失真度之間為一比例關係,故,晝面的複雜度可 定義成: 2 其中, 可得知 晝面的 晝 先前技 雜度, 同編碼 變化, ^ ^白勺 如果視 0_係該晝面的平均量化位階 位元率(實際消耗位元數)以 複雜度亦因此獲得。 面之複雜度視畫面的本質 術利用最近編碼畫面之複雜 並且為I_、P -及B~晝面分別 形式的影響,倘若視訊序列 則此方式可達到相當不錯的 目標位元配額與其實際消耗 訊序列的内容大幅變動,上 。在晝面編碼之後,將 及平均置化位階,而該 及其編碼的形式而定。 度來表示目前晝面之複 保持其複雜度以減輕不 的晝面其内容為平順的 效果。然而,由於畫面 位元數之間不相符合, 述方式並不適合用來達 1232684 五、發明說明(8) 成一致的視覺品質。 根據本發明,預先分析目前晝面的活動量(activhy) 以及先前編碼晝面的實際複雜度之間的關係’目前晝面之 複雜度將能藉此預估。活勳量係一種在畫面編碼後其位元 率和品質的量度嘗試,假設一整張晝面的全活動量A正比 於複雜度C,則 A:k 乂 其中5 k係比例常數。若k ’為活動量對複雜度之比 (activity-to-complexity ratio ? ACR),貝1J 全〉、舌動量 A 還 可以下列式子表示: A = ACRxrxg 根據本發明,可以從相同型態的先前編碼晝面來預估 ACR。觀念上,先在編碼一張晝面前計算其全活動量a 後,即將被編碼的晝面p⑴之複雜度巧以如下方式估瞀 rii) 其中,A(1)係晝0702-7606twF (nl); 90P132; Lin.ptd V. Description of the invention (6) Instantaneous bit rate of continuous picture, area = interval complexity and instantaneous ::: complexity of daytime surface, looseness of this group of heads: The amount is allocated to the current day. The upper target bit dispatcher will encode the 'quantization level actually used in the current picture according to the current full-digit updater of the current picture after encoding: the actual number of bits consumed and the amount of activity against the complex m · where, the value, Calculate the current screen degree = ratio and full activity of the current screen = the amount of activity on the diurnal surface versus the number of complex daily consumption levels and the current diurnal surface, which is inversely proportional to the current screen. In addition, the parameter updater also uses The average value is used to calculate the aforementioned statistical complexity. Based on the complexity of the I side, on the other hand, the present invention also discloses a momentary bit rate calculator, a quota suppression & bit dispatcher, including a bit rate calculator to receive the video sequence. 2 f , And a limit unit. The complexity of the current daytime surface of the instant code and the number of connections, which will be edited in the picture, will be updated based on the complexity: the system: complexity, and this group of connections will be calculated according to a function ^ 2 The function of the instantaneous complexity of the picture depends on the set of continuous ς = instantaneous bit rate, its complexity and instantaneous complexity. Quota rate, statistics, and bit rate! Based on the rate and instantaneous complexity, the goal: yuan yield f ::: given the picture screen 'and the limit unit is used to limit the target quota to a predetermined upper limit under. [Embodiment] In order to make the above-mentioned objects, features, and advantages of the present invention more comprehensible, (7 > V. Description of the invention will be described in detail as follows. A preferred embodiment is given below, in conjunction with the accompanying drawings: At the beginning, the prediction day and P-day surfaces must be decoded in the MPEG encoder. By calculating the root mean square error between the original day surface and the decoded day surface, the decoded day surface can be obtained. An objective measure of quality. Taking the root mean square error as the distortion of the decoding day, the complexity of the decoding day (complex) -C is defined as follows: C = r X d where 'r is the number required to encode the day surface. Bit rate, d is the degree of distortion of the diurnal plane of decoding. In theory, there is a proportional relationship between the average quantization level (quant 丨 zat 丨 〇nste ρ size) and the degree of distortion, so the complexity of the diurnal plane can be defined Cheng: 2 Among them, we can know the diurnal pre-technical complexity of the diurnal surface, which is the same as the coding change. ^ ^ If 0_ is the average quantization bit rate (the actual number of consumed bits) of the diurnal surface in terms of complexity It is also obtained as a result. The essence of the surface uses the complexity of the recently encoded picture and the effects of the I_, P-, and B ~ diurnal planes respectively. If the video sequence is used, this method can achieve a fairly good target bit quota and the content of the actual consumed signal sequence changes significantly. After the daytime encoding, the level is averaged and the level is determined, and the form of the encoding depends on the degree. The degree indicates that the current daytime complex maintains its complexity to reduce the effect of the daytime surface whose content is smooth. However, due to the inconsistency between the number of picture bits, the description method is not suitable for reaching 1232684 V. Invention Description (8) to achieve consistent visual quality. According to the present invention, the current day's surface activity (activhy) is analyzed in advance. And the actual complexity of the previously coded day-plane. 'The current day-plane complexity will be able to be estimated. The live metric is an attempt to measure the bit rate and quality of a picture after it has been coded, assuming a whole sheet The total amount of daytime activity A is proportional to the complexity C, then A: k 乂 where 5 k is a proportional constant. If k 'is the activity-to-complexity ratio? ACR, 1J total>, the tongue momentum A can also be expressed by the following formula: A = ACRxrxg According to the present invention, ACR can be estimated from the previously coded day-plane of the same type. Conceptually, first calculate its full activity before coding a day After measuring a, the complexity of the daytime surface p⑴ to be coded is estimated as follows: rii) where A (1) is the daytime
+ , i)的全活動量,⑼係同型態的,^ 編碼晝面其活動量對複雜度之比。以下,在μ 先刖 XA L· 4;® (i ) .. 爷'付 3虎或變數 的上軚表不該付號或變數係與即將被編碼的 關;同理,在符號或變數中的上標α_υ 則I面习 相關。 ⑺與先哥編碼晝面+, i) full activity, ⑼ is of the same type, ^ encodes the ratio of its activity to complexity on the day. In the following, before μμ XA L · 4; ® (i) .. Ye's pay 3 tigers or the upper part of the variable indicates that the pay number or variable is related to the code to be encoded; similarly, in the symbol or variable The superscript α_υ is related to I. ⑺ and Xiange coding day and night
1232684 五、發明說明(9) 估計的複雜度巧可用來為合適型態的畫面更新其複雜 度,所以三種型態的晝面其複雜度Ci、Cp和CB將分別儲存以 利後續之處理。一組連續畫面之瞬間複雜度 (instantaneous complexity)亦以如下方式更新: INST 一 C = X C: + iVp X X。方 其中,N〗、NP和NB分別是該組連續晝面中卜、P -及B_畫面之 數目,而且此處所指的一組連續畫面至少包含一組G0P。 除了 G、CP和CB之外,亦分別存放I _、P -及B-晝面之複雜度 統計量sq、SCP和SCB,並且取目前晝面之複雜度巧來更新 同型態晝面之複雜度統計量,則一組連續晝面之統計複雜 度為:1232684 V. Description of the invention (9) The estimated complexity can be used to update the complexity of the appropriate types of pictures, so the complexity Ci, Cp, and CB of the three types of day surfaces will be stored separately for subsequent processing. The instantaneous complexity of a set of continuous pictures is also updated as follows: INST-C = X C: + iVp X X. Where N, NP, and NB are the numbers of Bu, P-, and B_ pictures in the group of continuous diurnal planes, respectively, and a group of continuous pictures referred to here includes at least one group of GOP. In addition to G, CP, and CB, I_, P-, and B- daytime complex statistics sq, SCP, and SCB are also stored, and the current daytime complexity is used to update the isomorphic daytime surfaces. Complexity statistics, the statistical complexity of a set of continuous diurnal surfaces is:
STAT^C^Nj X SCj ^Npx SCP ^NBx SCB 若是最大、最小位元率已根據每組連續晝面期望達到的編 碼位元率而事先決定,如此,一組連續畫面之瞬間位元率 可由下列式子計算而得: INST 一 R=ER^an·STAT ^ C ^ Nj X SCj ^ Npx SCP ^ NBx SCB If the maximum and minimum bit rates have been determined in advance according to the encoding bit rate expected for each group of continuous day-to-day faces, so the instant bit rate of a group of continuous pictures can be determined by It is calculated by the following formula: INST-R = ER ^ an ·
/ (INST C 1 + SF' -=—-1 \ [static JJ 其中,ERmean係本組連續晝面之既定平均位元率,sf係一縮鲁 放因子,其範圍在0到1之間;瞬間位元率INST —R還須進— 步地限制在每組連續晝面之既定最大、最小位元率之間: ERk<INST_R<ER_/ (INST C 1 + SF '-= —- 1 \ [static JJ, where ERmean is the predetermined average bit rate of the continuous day and day in this group, and sf is a contraction factor, which ranges from 0 to 1; Instantaneous bit rate INST —R must also advance — is further limited to the set maximum and minimum bit rate of each group of consecutive day and time: ERk < INST_R < ER_
一旦瞬間複雜度孤ST一C和瞬間位元率INST 則目前晝面P(1)之目標位元額度TB⑴為 ΤΒ(ί)=Once the instantaneous complexity ST-C and the instantaneous bit rate INST, the target bit quota TB⑴ of the current day P (1) is ΤΒ (ί) =
INS? CINS? C
xINST R 由上式可知,目標位元額度以⑴係與複雜度⑦以及瞬間位 兀率INST一R成正比,而與瞬間複雜度成反比。 所有的MPEG-2資料位元流均必須遵循MpEG—2標準的 VBV規則,分配的目標額度必須受限以使νβν缓衝器不會滿 溢(over flow)或匱乏(under flow)。當目前晝面之目標位 =額度決定好之後,便可利用許多不同型態的影像編碼 益,像是MPEG-2標準描述的測試原型5(Test M〇del 5), 根據目標位元額度來完成晝面的壓縮。 目岫晝面在完成編碼之後,其實際的複雜度將可得 知,而目前畫面的全活動量和實際複雜度之間的關係, ACR(1)可從以下計算得到 ACR(i) =_ AQ{i) x UB[i) 其中,AfRM與全活動量Α(υ成正比,而與目前晝面的實際 消耗位7L數UB(1)以及平均量化位階AQ⑴成反比。此ACR(n可 拿來預估下一張同型態的畫面之複雜度。ACR⑴可以和 ACR(1 υ做線性結合來避免受到那些富含雜訊畫面的影響。 士發明的單次(311^1卜1^3)視訊編碼可藉由第3圖之 車乂佳只施例並配合第4圖之操作流程圖來解釋。如第3圖所 1232684 五、發明說明(11) :声ϊ :態?像壓縮裝置3 0 0包括1動估算單元31 〇 複 雜度估計哭^ α切俗异早兀diu、一 34〇 jv . 、一目裇位兀为派器330、一參數更新器 列相1 像編碼器35〇。移動估算單元接收視訊序 執rt組連續畫面中即將被編瑪的目前畫面P⑴,用來在 量AC异的區塊匹配時’計算此目前晝⑴之全活動 同型育vm基於全活動量a(i)以及這組連續晝面中 ";、的2剛編碼晝面之ACR(1-〗),複雜度估計器32 0可用 來估计目前晝面之複雜度°2 (步驟S420 )。目標位元分派器 ,二t的^雜度。2更新本組連續晝面其瞬間複雜度 新哭二驟S43〇),同時,目標位元分派器33 0從參數更 這:連病壹收面該」且連續畫面之統計複雜度STAT—C,並根據 "二t 既定平均位元率別^、統計複雜度 元率INST—R(步驟S44〇) Γ片开該組連績晝面之瞬間位 ⑺目払位το分派器33〇接著以瞬間位元率、複雜 =目以瞬面:複雜度·-C為基礎將目標位元額度TB⑴分 標;二而(:=:因此,影像編物輪 這些$化位階係作為晝面p⑴ t二位白 的編碼之用。影傻給石工間區域·大區塊(MB) 晝面P⑴(步驟S4e(〇。在^·t,據這些,化位階編碼目前 3 5 0將查® P⑴夕每 刖旦面編碼完成後,影像編碼器 階之平旦均值AQ⑴:Γ消耗位元數顺)以及實際所用量化位 一 回報給务數更新器340。然後以全活動量From the above formula, xINST R shows that the target bit quota is proportional to the system complexity and the instantaneous bit rate INST-R, and is inversely proportional to the instantaneous complexity. All MPEG-2 data bit streams must follow the VBV rules of the MpEG-2 standard, and the allocated target quota must be limited so that the νβν buffer does not overflow or under flow. After the current day's target position = quota is determined, many different types of image coding benefits can be used, such as the Test Prototype 5 (Test Model 5) described in the MPEG-2 standard, based on the target bit quota. Complete the day and night compression. After the coding of the daytime plane is completed, its actual complexity will be known, and the relationship between the full activity of the current picture and the actual complexity, ACR (1) can be calculated from the following ACR (i) = _ AQ {i) x UB [i) where AfRM is directly proportional to the full activity A (υ) and inversely proportional to the current actual consumption level of 7L UB (1) and the average quantization level AQ 位. This ACR (n 可It is used to estimate the complexity of the next picture of the same type. ACR⑴ can be linearly combined with ACR (1 υ to avoid being affected by those pictures that are rich in noise. A single shot (311 ^ 1b1 ^ 3) The video coding can be explained by the example of the car in Figure 3 and the operation flowchart of Figure 4. As shown in Figure 1232684 V. Description of the invention (11): Sound: State? Image compression The device 3 0 0 includes a motion estimation unit 31 〇 complexity estimation crying ^ α vulgarity early diu, 34 0jv., A single head is a dispatcher 330, a parameter updater phase 1 image encoder 35 〇. The mobile estimation unit receives the current picture P⑴ in the continuous pictures of the video sequence rt group to be edited, and is used to calculate the number of blocks when the AC is different. Based on the full activity amount a (i) and the 2 just-coded ACR (1-) in this group of continuous diurnal planes, the complexity estimator 32 0 is available. To estimate the current complexity of the daytime surface ° 2 (step S420). The target bit dispatcher, the complexity of the two t. 2 Update the instantaneous complexity of this group of continuous daytime surfaces (new step S43). At the same time, the target The bit dispatcher 33 0 changes it from the parameters: the statistical complexity of the continuous picture is STAT-C, and according to the "average bit rate of the set average ^, the statistical complexity element rate INST- R (step S44) Γ opens the instantaneous position of this group of consecutive day-to-day positions οο dispatcher 33 〇 Then based on the instantaneous bit rate, complexity = instantaneous plane: complexity--C based on the target Bit quota TB⑴ sub-standard; Second, (: =: Therefore, these $ gradation levels of the image editing wheel are used as the coding of the daytime p⑴t two-dimensional white. The shadow silly gives the masonry area and large block (MB) day The plane P⑴ (step S4e (0. At ^ · t, according to these, the level coding is currently 3 50 will be checked. After the plane coding is completed, the image encoder stage Pingdan mean AQ⑴: Γ consumption according to the number of bits) and actually used to quantify a return to the position number 340. The updater works then in full activity
0702-7606twF(nl) ; 90P132 ; Lin.ptd 第16頁 1232684 ----- 五、發明說明(12) A(1)、實際消耗位元數UB⑴和實^所3 基礎,參數更新器34 0為;之平均值 對複雜度之比ACR⑴(步驟S4 70 )。旦 计异其活動量 等效上,第3圖之較佳實施例 軟體來實現。根據本發明,移動估曾考篁以硬體以及/或是 其他元件可以管線(pipeline)=式二=凡310和第3圖中的 320開始計算目前晝面的複雜度 ,在硬〃雜度估計器 須先完成即將被編碼的目前晝面::動旦估异早X31 〇必 :向量之估算,並且,當移動估算::之计鼻以及移 準備時,複雜度估舛哭q 9 #甘从 3 1 0為下張畫面作 冰 ^ ^ 3 2 0和其他元件仍正忙於办士仏上 與目前畫面相關的運作。接下夾 、凡成所有 中的每個元件。 接下末將砰細地描述較佳實施例 私動估异單元3 1 〇的主要的一 模式夾绝m π▲ 兩厌疋用何種預測 6义^、,扁馬一張晝面裡的每個MB,如果必要的話,亦進 後的移動預測’還可從區塊匹配運算中提取書‘ 動I的貧訊。首先,計算每個MB之自身活動量(intr~a〜 $CtlVlty) ’將一個MB中4個8x8亮度值區塊之像素強度以 m’n ’^0’···’7,11 = 〇,.一,7來表示,並且每個8\8區塊其 平句值為F ’則母個MB之内活動量IntraAct ··0702-7606twF (nl); 90P132; Lin.ptd Page 16 1232684 ----- V. Description of the invention (12) A (1), actual number of consumed bits UB⑴ and the real base 3, parameter updater 34 0 is the ratio of the average value to the complexity ACR⑴ (step S4 70). Once the amount of activity is equivalent, the software of the preferred embodiment of FIG. 3 is implemented. According to the present invention, mobile estimation has tested whether the hardware and / or other components can be used to calculate pipeline = Equation 2 = Fan 310 and 320 in Figure 3, and calculate the current complexity of the daytime surface. The estimator must first complete the current day-to-day encoding :: Dynamic estimation early X31 〇 Must: the estimation of the vector, and, when the motion estimation :: the nose and the preparation of the movement, the complexity estimation cries q 9 # Gan made ice for the next picture from 3 1 0 ^ ^ 3 2 0 and other components are still busy with the operations related to the current picture on the office. Connect each element in the clip, Fan Cheng All. Next, the main mode of the private movement estimation unit 3 1 〇 of the preferred embodiment will be described in detail. What kind of prediction is used by the two dissatisfiers? For each MB, if necessary, the post-movement prediction 'can also extract the book' from the block matching operation. First, calculate the amount of each MB ’s own activity (intr ~ a ~ $ CtlVlty) 'Set the pixel intensity of four 8x8 luminance blocks in a MB to m'n' ^ 0 '...', 11 = 〇 ,. 一 , 7 , and the flat sentence value of each 8 \ 8 block is F ', then the amount of activity within the parent MB is IntraAct ··
IntmAct = ^pk 其中IntmAct = ^ pk where
k^O 第17頁 0702-7606twF(nl) ; 90P132 ; Lin.ptd 1232684 五、發明說明(13) °h = J Σ ni^??v2^yfk ^ O Page 17 0702-7606twF (nl); 90P132; Lin.ptd 1232684 V. Description of the invention (13) ° h = J Σ ni ^ ?? v2 ^ yf
Vwf〇k*0 倘若需要較低的計算複雜度,可以相對P的絕對差值代 替··Vwf〇k * 0 If a lower computational complexity is required, it can be replaced with the absolute difference of P ...
IntraAct = λ=0 其中 由於I-晝面中的MB僅有一種模式:;[模式,因此IntraAct 即為I -畫面的每個MB之活動量。 如果晝面為P-或B-晝面,則需進行移動估算。最常使 用來發現最佳移動向量之技術係區塊匹配。對非自身編碼 (nonintra coding)而言(如p一和B一晝面),利用將失真標 準如變異數或失真絕對值和減至最低,來選擇向前、向 後、雙向的預測或不需移動補償。一旦卜或B—畫面中每個 MB之MB模式決定,在每個移動補償的差值Μβ中4個8以區塊 之變異數亦可求得,移動補償的差值MB係待處理〇和預估 MB間像素對像素之差;失真絕對值和通常具有較佳的計瞀 效率,因此可用來取代變異數。將4個8乂8區塊之變里數2 =絕對值和相加以求得非内編碼畫面中每娜之相互活 ,(:inter-actlV1ty),InterAct,然後把非内編碼畫面 中母個MB之IntraAct及其InterAct拿來作比較,以判斷 i-j 第18頁 0702-7606twF(nl) ; 90P132 ; Lm.ptd 1232684 五、發明說明(14) I^iterAct是否較小,若是,則以InterAct作為該Μβ之活動 量,並以相互模式(inter-mode)編碼該mb ;否則以IntraAct = λ = 0 where MB in I-day has only one mode: [mode, so IntraAct is the activity of each MB in I-picture. If the diurnal surface is a P- or B-diurnal surface, a motion estimation is required. The technique most commonly used to find the best motion vector is block matching. For nonintra coding (such as p- and B-days), use the minimum and minimum distortion standards such as the number of variations or the absolute value of distortion to choose forward, backward, or bidirectional prediction or not. Motion compensation. Once B or B—The MB mode of each MB in the picture is determined, 4 8 blocks of variation values in each motion compensation difference Mβ can also be obtained. The motion compensation difference MB is to be processed. 0 and The pixel-to-pixel difference between MBs is estimated; the absolute value of distortion and usually have better computational efficiency, so it can be used instead of the number of variations. Change the number of 4 8 乂 8 blocks to 2 = absolute value and add up to find each other's live in non-inner coded picture, (: inter-actlV1ty), InterAct, and then put the mother in the non-inner coded picture MB's IntraAct and its InterAct are compared to judge ij page 18 0702-7606twF (nl); 90P132; Lm.ptd 1232684 V. Description of the invention (14) I ^ iterAct is small, if yes, then InterAct is used as The amount of activity of the Mβ, and the mb is coded in inter-mode; otherwise,
IntraAct作為該MB之活動量,並以自身模式(intra_m〇d 編碼該MB。最後,對目前的ί _、p_或R佥; 7曰月J扪丄尸或晝面,將所有MB之 活動篁相加而得全活動量Au),移動估算單元31〇再把全活 動量A⑴傳送給複雜度估計器32〇以及參數更新器。 接著,複雜度估計器32 0為某種型態的目前晝舒⑴估 异其複雜度,且依照三播壹而剂At p] 〜種旦面1悲,引進加權係數至預估 的複雜度巧。因為B〜奎而p 丁处、士 m + 、 故可減少B-書面之加預估其他的晝面, 保留較多的:元給卜和較,位元給Β-晝面而 產生最多的位元,因此p :,一般而吕,編碼I -畫面會 加權係數。複雜度估补:晝面之加權二係數又小於卜畫面之 雜度q、cP或cB三者其φ m復 如下方式估算: 中之一,目前畫面P⑴之複雜度e以 if (I-晝面) 1se if (P- ACR^l) 晝 面) 'Cp' ACR(^l) lse if (B- 晝 面)IntraAct is the activity amount of the MB, and encodes the MB in its own mode (intra_m0d. Finally, for the current _, p_, or R 佥; July 7th, J 扪 丄 corpse or daytime, all MB activities篁 is added to obtain the full activity amount (Au), and the motion estimation unit (31) transmits the full activity amount (A) to the complexity estimator (32) and the parameter updater. Next, Complexity Estimator 320 is a certain type of day comfort ⑴ estimates vary its complexity, and the agent in accordance with three multicast One At p] ~ 1 sad face deniers species, the introduction of weighting coefficients to estimate the complexity of the current Coincidentally. Because B ~ Quer and p Ding, Shi m +, it can reduce the B-written plus estimate of the other day surface, and keep more: Yuan to Bu and compare, bit to the B- day surface and produce the most Bits, so p :, but generally, encoding I-pictures will weight coefficients. Computation of complexity: The weighted two coefficients of the daytime surface are less than the miscellaneousness q, cP, or cB of the picture. The φ m is estimated as follows: One, the complexity e of the current picture P⑴ is given by if (I-day Surface) 1se if (P- ACR ^ l) day surface) 'Cp' ACR (^ l) lse if (B- day surface)
1232684 發明說明(15)1232684 Description of the invention (15)
A(i} ACR(fl) 其中,υ及^^厂係一組連續畫面中,ι、ρ 和Β型態的先w編螞畫面各自的活動量對複雜度之比。這 和^Ci^2l;存放在參數更新器34 0之中, 而複雜度估計器32 0會為適當型態的先前畫面讀取對應的 活動量對複雜度之比。Κι、1?及&分別係卜、卜和β-晝面之 加權係數,其範圍一般是在〇到i · 〇之間。至於丨_晝面,較 佳實施例可以採用1 = i, 0。 一A (i) ACR (fl) where υ and ^^ are a series of continuous pictures, and the ratio of the activity amount to the complexity of the first w-edited pictures of the ι, ρ, and B types. This is ^ Ci ^ 2l; is stored in the parameter updater 34 0, and the complexity estimator 320 reads the ratio of the corresponding amount of activity to the complexity for the previous picture of the appropriate type. K1, 1? And & The weighting coefficient of Bu and β-day plane is generally in the range of 0 to i · 〇. As for the 丨 _ day plane, the preferred embodiment can use 1 = i, 0.
目標位兀分派器330可為一組連續畫面中的三種型能 :I-晝面、P-晝面及B-畫面分別儲存其複雜度匕、複雜,度The target position dispatcher 330 can be three types of capabilities in a set of continuous pictures: I-day face, P-day face, and B-picture respectively store their complexity, complexity, and degree.
Cp與複雜度CB。<t故到C2.之時,視目前畫面p( , 母次僅更新複雜度C〗、CP、CB三者其中之一,如下. if (I-晝面) el se if (P-畫面)Cp and complexity CB. < t So when C2., depending on the current picture p (, the parent and child only update one of the complexity C〗, CP, CB, as follows. if (I-day surface) el se if (P-screen )
GO else if (B-晝面)GO else if (B-Day Face)
如第5圖所示之較佳實施例’目標位元分派器33〇係由瞬間 位兀率計算器510、配額計算器5 20以及限度單元53〇所組 成。瞬間位元率計算器510為該組連續畫面更新立瞬間複The preferred embodiment 'target bit dispatcher 33o' shown in Fig. 5 is composed of a momentary bit rate calculator 510, a quota calculator 520, and a limit unit 53. Instant bit rate calculator 510 instantly updates the set of continuous pictures
1232684 五、發明說明(16) 如下1232684 V. Description of the invention (16) is as follows
雜度 JJVST_CMiscellaneous JJVST_C
IN^T^C^NIxCI-l· NpxCp^NBxCB 其中’ N!、NP和nb分別是一組連續晝面中卜、p一及B—畫面之 數目。在此同時,瞬間位元率計算器5 1 〇還從參數更新器 340讀取這組連續晝面之統計複雜度STAT—C,並且依據下 式計异一組連續晝面之瞬間位元率I nst_r :IN ^ T ^ C ^ NIxCI-l. NpxCp ^ NBxCB where ‘N !, NP, and nb are the numbers of p, p, and B—pictures in a set of continuous daylight planes, respectively. At the same time, the instantaneous bit rate calculator 5 1 0 also reads the statistical complexity of this set of continuous day-to-day surfaces from the parameter updater 340, and calculates the instantaneous bit rate of a set of continuous day-to-day surfaces according to the following formula I nst_r:
(INS7_C KSTA7_C \ 其中’ ERmean係本組連續畫面之既定平均位元率,gp係一縮 放因子,其範圍在〇到1之間;而瞬間位元率INST—R最好限 制在這組連續晝面的最大、最小位元率ER_^及之間。 配額計算器52 0接著分配目標位元額度ΤΒ⑴給目前書面 Ρ(1),而目標位元額度ΤΒ⑴係根據 | ΤΒ(ϋ = ^est Jnst c(INS7_C KSTA7_C \ where 'ERmean is the predetermined average bit rate of the continuous picture in this group, gp is a scaling factor ranging from 0 to 1; and the instantaneous bit rate INST-R is preferably limited to this group of continuous days The maximum bit rate and minimum bit rate ER_ ^ between the two. The quota calculator 52 0 then allocates the target bit quota TB1 to the current written P (1), and the target bit quota TB is based on | ΤΒ (ϋ = ^ est Jnst c
XINST R 但是’貫際的消耗位元數並不會和期望的位元額度剛 等,因此發展一種回授策略來使晝面實際消耗的位元』 近於目標位元額度’在較佳實施例中,編碼至目前為止、 超用位元量,可由正在編碼的目前晝面攤還一部份·· 的XINST R However, 'the number of consumed bits will not be equal to the expected bit quota, so develop a feedback strategy to make the actual bits consumed by the day surface "close to the target bit quota" is better implemented In the example, the amount of overused bits that have been encoded so far can be partly amortized from the current day being encoded ...
TB(i) = TB(i)-ARxEB 其中,EB係參數更新器3 40傳來的超用位元量,AR則係艮 定的攤還率,其範圍一般是在〇· 05到0·2之間。按照前5TB (i) = TB (i) -ARxEB Among them, the amount of overused bits transmitted by the EB-based parameter updater 3 40, and AR is a fixed amortization rate, which generally ranges from 0.05 to 0. Between 2. Follow top 5
1232684 五、發明說明(17) 的式子,目標位元額度TB⑴係與目前畫面之複雜度θΐ以及 —組連續晝面之瞬間位元率IN S T 一R成正比,而與一組連續 晝面之瞬間複雜度成反比。限度單元5 3 0則能用來 调整目標位元額度ΤΒ(1)以符合VBV規範,因此事先定下了額 度的上限(ILbound)及下限(L-bound)。就固定位元率(CBR )的操作而言,分配給一張畫面的目標位元額度須使VBV緩 衝器不會滿溢或匱乏,所以限度單元5 3 0將目標位元額度 丁 B⑴限制在上、下限範圍内·· if ( TB⑴〉U—bound) then TB⑴=U—bound if ( TB(1) < L—bound) then TB⑴=L—bound 若是可變位元率(VBR),則只要防止VBV匱乏即可,故: if ( TB(1) > U — bound) then TB⑴=U-bound 然後’目標位元額度TB⑴會被傳送到參數更新器3 40以及影 像編碼器3 5 0。 回到第3圖’影像編碼器3 5 0透過訊號線31 2從移動估 异單元3 1 0接收晝面資料以及每個μ B的移動向量和Μ B模式 。為了儘可能的使晝面達到最佳品質,影像編碼器3 5 〇依 據目標位元額度ΤΒ(1)來決定目前晝面的每個Μβ編碼所用之 量化位階’以這些資料為基礎,影像編碼器3 5 〇開始對目 前畫面Ρ⑴進行編碼且輸出編碼後的資料⑶。在目前晝面 Ρ(1)編碼之後,影像編碼器3 5 〇將所用的量化位階做平均並 且計算實際消耗位元數UB⑴,再回報目前畫面的ϋΒ⑴以及 貝際所用量化位階之平均值AQ⑴給參數更新器340。1232684 V. Formula of invention description (17), the target bit quota TB⑴ is proportional to the complexity θΐ of the current picture and the instantaneous bit rate IN ST-R of a group of continuous day-planes, and is proportional to a group of continuous day-planes The instantaneous complexity is inversely proportional. The limit unit 5 3 0 can be used to adjust the target bit quota TB (1) to comply with the VBV specification. Therefore, the upper limit (ILbound) and the lower limit (L-bound) of the quota are set in advance. As for the operation of a fixed bit rate (CBR), the target bit quota allocated to a picture must be such that the VBV buffer will not overflow or be scarce. Therefore, the limit unit 5 30 limits the target bit quota D B to Within the upper and lower limits ... if (TB⑴> U-bound) then TB⑴ = U-bound if (TB (1) < L-bound) then TB⑴ = L-bound If it is a variable bit rate (VBR), then You only need to prevent VBV scarcity, so: if (TB (1) > U — bound) then TB⑴ = U-bound Then the 'target bit quota TB⑴ will be transmitted to the parameter updater 3 40 and the image encoder 3 5 0 . Returning to FIG. 3 ', the image encoder 3 50 receives the day-to-day data from the motion estimation unit 3 1 0 through the signal line 31 2 and the motion vector and MB mode of each μB. In order to achieve the best quality of the daytime plane, the image encoder 3 5 〇 determines the quantization level used for each Mβ encoding of the current dayplane based on the target bit quota TB (1). Based on these data, the image coding The device 3 50 starts to encode the current picture PZ and outputs the encoded data ⑶. After the current day plane P (1) encoding, the image encoder 3 5 0 averages the quantization levels used and calculates the actual number of consumed bits UB⑴, and then returns the current picture ϋΒ⑴ and the average value of the quantization levels used by Bayesian AQ⑴ to Parameter updater 340.
1232684 五、發明說明(18) 晝面:p數二會面為八一:連續晝面中之三種型態:1_ ^〇Γ 一 刀別儲存其複雜度統計量sq、SCp 興84,並且取晝面之複 — 11232684 V. Description of the invention (18) Day surface: p-number two meeting is eighty-one: three types of continuous day surface: 1_ ^ 〇Γ One store the complexity statistics sq, SCp Xing 84, and take the day surface Complex — 1
〜旦面之稷雜度統計量,如下· 但I谿T丨j I 權係數。換言之,’'二面、卜晝面及β—晝面之既定加 SCI、SC «C口# ^又據耵晝面Ρ⑴的型態,複雜度統計量 I 或SCB母次只有其中一個會被更新: if (I-晝面) L· 松 SC: ^ SCI" (WI ^1)+ w2 else if (p—畫面) scr ^ SCP ' (WP^}) + Wp else if (b-畫面)The statistics of the impurity degree of the surface are as follows: · But the coefficient of the weight of I xi T 丨 j I. In other words, according to the type of '' dihedral surface, diurnal surface, and β-diurnal surface plus SCI, SC «C 口 # ^ and the diurnal surface P⑴, only one of the complexity statistics I or SCB parent is Update: if (I-Day) L. Song SC: ^ SCI " (WI ^ 1) + w2 else if (p—picture) scr ^ SCP '(WP ^)) + Wp else if (b-picture)
scr =: " (WB ^ jj + C^J 一般是在200至2000之間。參數更 程式計算一組連續晝面之統計複scr =: " (WB ^ jj + C ^ J is generally between 200 and 2000. The parameter is updated. The program calculates the statistical complex of a set of continuous diurnal surfaces.
其中,Wi、Wp和WB的範圍 新器3 4 0接著以下列的方 雜度STAT_C, STATIC ^ Nj X ^ + ΛΓΡ X + NB x SC‘ 1232684Among them, the range of Wi, Wp, and WB. The new device 3 4 0 then follows the degree of heterogeneity STAT_C, STATIC ^ Nj X ^ + ΛΓΡ X + NB x SC ’1232684
五、發明說明(19) 其中,乂、NP和NB分別是每組連續畫面中卜、p-及B-畫面之 數目。晝面P(1)之目標位元額度^⑴以及實際消耗位元婁 UB(1)間的差距’會由參數更新器wo做累計以便在畫面P 編碼之後付到超用位元量E B : EBx(^L-AR)^UQa) _ Τβ〇) 其中’ AR係既定的攤還率。因此,活動量和實際衩雜度 間的關係,ACR(1),可從以下計算得到 ACRU)= A(i)— AQU) x 這個剛編碼完成晝面P(1)所屬之ACR⑴可用來預估下 張一 / 型態畫面之複雜度。為使ACR⑴不會受到富含雜訊畫面的影 響,較佳實施例利用了 ACR⑴和ACRU-1)的線性組合,例 如: if (I-晝面) ACR}0 = ACR(^l) x (l - CW)^ ACR(i) x CW else if (P—晝面)V. Description of the invention (19) where 乂, NP and NB are the number of Bu, p- and B-pictures in each group of continuous pictures. The difference between the target bit quota ^ ⑴ of the day plane P (1) and the actual consumed bit Lou UB (1) will be accumulated by the parameter updater wo in order to pay the overused bit amount EB after the picture P is encoded: EBx (^ L-AR) ^ UQa) _ Tβ〇) where 'AR is the established amortization rate. Therefore, the relationship between the amount of activity and the actual impurity level, ACR (1), can be calculated from the following ACRU) = A (i) —AQU) x The ACR⑴ to which the daytime plane P (1) belongs can be used to predict Estimate the complexity of the next Zhang Yi / style picture. To prevent ACR⑴ from being affected by noise-rich pictures, the preferred embodiment uses a linear combination of ACR⑴ and ACRU-1), for example: if (I-day) ACR} 0 = ACR (^ l) x ( l-CW) ^ ACR (i) x CW else if (P—day surface)
ACR(;} = ACR(;-l) x (1 ~ CW)^ ACR(i) x CW el se i f (B—晝面)ACR (;) = ACR (;-l) x (1 ~ CW) ^ ACR (i) x CW el se i f (B—day surface)
ACR(^ = ACRTl) X fl ~ cw)+ ACR(i) x CW 其中,CW係線性組合既定之加權係數。更新過的、 及〃會被傳送到複雜度估計器3 2 0,以便為下一 張適當型態的晝面估算其複雜度,此外,超用位元量EB以ACR (^ = ACRTl) X fl ~ cw) + ACR (i) x CW Among them, CW is a predetermined weighting coefficient of linear combination. The updated and 〃 are transmitted to the complexity estimator 3 2 0 in order to estimate its complexity for the next day of the day of the appropriate type. In addition,
0702-7606twF(nl) ; 90P132 ; Lin.ptd 第24頁 1232684 五、發明說明(20) 及統計複雜度STAT_C則送至目標位元分派器330作為位元 分配之用。 雖然本發明已以一具體實施例揭露如上,然其僅為了 易於說明本發明之技術内容,而並非將本發明狹義地限定 於該實施例,任何熟習此技藝者,在不脫離本發明之精神 和範圍内,當可作些許之更動與潤飾,因此本發明之保護 範圍當視後附之申請專利範圍所界定者為準。0702-7606twF (nl); 90P132; Lin.ptd page 24 1232684 V. Description of the invention (20) and statistical complexity STAT_C are sent to the target bit dispatcher 330 for bit allocation. Although the present invention has been disclosed as above with a specific embodiment, it is only for easy explanation of the technical content of the present invention, rather than narrowly limiting the present invention to this embodiment. Any person skilled in the art will not depart from the spirit of the present invention. Within the scope and scope, some modifications and retouching can be made, so the scope of protection of the present invention shall be determined by the scope of the appended patent scope.
0702-7606twF(nl) ; 90P132 ; Lm.ptd 第25頁 1232684 圖式簡單說明 第1圖係以顯示的順序呈現典型的畫面群組(GOP ); 第2圖係MPEG之大區塊; 第3圖係本發明的可變位元率控制裝置其較佳實施例 之方塊示意圖; 第4圖係本發明之操作流程圖;以及 第5圖係第3圖中的目標位元分派器之方塊示意圖。 【符號說明】 3 0 0〜動態影像壓縮裝置; 31 0〜移動估算單元; 31 2〜訊號線; 32 0〜複雜度估計器; 3 3 0〜目標位元分派器; 34 0〜參數更新器; 3 5 0〜影像編碼器; 5 1 0〜瞬間位元率計算器; 5 2 0〜配額計算器; 5 3 0〜限度單元; P⑴〜晝面; A(1)〜全活動量; β〜預估複雜度; ACR(1)〜活動量對複雜度之比; EB〜超用位元量;0702-7606twF (nl); 90P132; Lm.ptd Page 25 1232684 Brief description of the diagram Figure 1 shows a typical group of pictures (GOP) in the order of display; Figure 2 is a large block of MPEG; Figure 3 Figure 4 is a block diagram of a preferred embodiment of the variable bit rate control device of the present invention; Figure 4 is a flowchart of the operation of the present invention; and Figure 5 is a block diagram of the target bit dispatcher in Figure 3 . [Symbol description] 3 0 0 ~ moving image compression device; 3 0 0 ~ motion estimation unit; 3 2 2 ~ signal line; 3 2 0 ~ complexity estimator; 3 3 0 ~ target bit dispatcher; 3 0 0 ~ parameter updater ; 3 5 0 ~ image encoder; 5 1 0 ~ instantaneous bit rate calculator; 5 2 0 ~ quota calculator; 5 3 0 ~ limit unit; P⑴ ~ day surface; A (1) ~ full activity amount; β ~ Estimated complexity; ACR (1) ~ Ratio of activity to complexity; EB ~ Over-utilized bit amount;
0702-7606twF(nl) ; 90P132 ; Lin.ptd 第26頁 1232684 圖式簡單說明 STAT — C〜統計複雜度; ΤΒ(υ〜目標位元額度; AQ(1)〜實際所用量化位階之平均值; UB(1)〜實際消耗位元數; CD〜編碼後的資料; INST_R〜瞬間位元率; INST_C〜瞬間複雜度。0702-7606twF (nl); 90P132; Lin.ptd page 26 1232684 The diagram briefly illustrates STAT-C ~ statistical complexity; TB (υ ~ target bit quota; AQ (1) ~ average of the quantization level actually used; UB (1) ~ the actual number of bits consumed; CD ~ encoded data; INST_R ~ instant bit rate; INST_C ~ instant complexity.
0702-7606twF(nl) ; 90P132 i Lm.ptd 第 27 頁0702-7606twF (nl); 90P132 i Lm.ptd page 27
Claims (1)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW92120340A TWI232684B (en) | 2003-07-25 | 2003-07-25 | Apparatus for variable bit rate control in video compression and target bit allocator thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW92120340A TWI232684B (en) | 2003-07-25 | 2003-07-25 | Apparatus for variable bit rate control in video compression and target bit allocator thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
TW200505244A TW200505244A (en) | 2005-02-01 |
TWI232684B true TWI232684B (en) | 2005-05-11 |
Family
ID=36320106
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW92120340A TWI232684B (en) | 2003-07-25 | 2003-07-25 | Apparatus for variable bit rate control in video compression and target bit allocator thereof |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI232684B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI421798B (en) * | 2010-04-27 | 2014-01-01 | Novatek Microelectronics Corp | Method and apparatus for image compression bit rate control |
-
2003
- 2003-07-25 TW TW92120340A patent/TWI232684B/en not_active IP Right Cessation
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI421798B (en) * | 2010-04-27 | 2014-01-01 | Novatek Microelectronics Corp | Method and apparatus for image compression bit rate control |
Also Published As
Publication number | Publication date |
---|---|
TW200505244A (en) | 2005-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
TWI304297B (en) | Video encoding methods and systems with frame-layer rate control | |
TW404132B (en) | A real-time single pass variable bit rate control strategy and encoder | |
US9197912B2 (en) | Content classification for multimedia processing | |
US8406297B2 (en) | System and method for bit-allocation in video coding | |
TW200533206A (en) | A video encoder and method of video encoding therefor | |
WO2012076646A1 (en) | High-dynamic range video tone mapping | |
CN102281446B (en) | Visual-perception-characteristic-based quantification method in distributed video coding | |
JPH0974565A (en) | Controller for video signal coding system | |
CN107155112A (en) | A kind of compressed sensing method for processing video frequency for assuming prediction more | |
Liu et al. | A novel rate control scheme for panoramic video coding | |
Ramanathan et al. | Rate-distortion analysis for light field coding and streaming | |
TW201018244A (en) | Flow control method of real-time video encoding chip | |
Flack et al. | Low-bandwidth stereoscopic image encoding and transmission | |
TWI232684B (en) | Apparatus for variable bit rate control in video compression and target bit allocator thereof | |
US20050195901A1 (en) | Video compression method optimized for low-power decompression platforms | |
Wang et al. | View synthesis distortion model based frame level rate control optimization for multiview depth video coding | |
TWI498005B (en) | Methods for coding and decoding a block of picture data, devices for coding and decoding implementing said methods | |
CN103248890A (en) | Source data acquisition and compression method for wireless projection | |
US20130077674A1 (en) | Method and apparatus for encoding moving picture | |
Xin et al. | Bit-allocation for transcoding pre-encoded video streams | |
TWI221739B (en) | Method and apparatus for moving picture compression rate control using bit allocation with initial quantization step size estimation at picture level | |
EP1921866A2 (en) | Content classification for multimedia processing | |
TWI226804B (en) | Apparatus for constant quality rate control in video compression and target bit allocator thereof | |
CN103391436B (en) | A kind of method for controlling three-dimensional video code rates based on binocular vision characteristic | |
Cheung et al. | Graphics-to-video encoding for 3G mobile game viewer multicast using depth values |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
MM4A | Annulment or lapse of patent due to non-payment of fees |