200922334 九、發明說明： 【發明所屬之技術領域】 本發明為提供一種小波視訊編碼，採用離散小波轉換和移動估測方法 來達到壓縮效果，而達到運算複雜度和所需記憶體均較傳統方法來的少之 適用於小波視訊編碼之低複雜度頻域移動估測方法。 【先前技術】 按’伴隨著資訊的快速成長，影音的資料量也越來越龐大，應用的領 域也有巨大的改變，所轉像魏的重要性也賊來越麵，在網路的蓬 勃發展之下，許？資繼縮及熱傳輪麟也跟著被人們討論與研究，經 過夕年的發展’廳技術也已有轉顯著的妙，國社已有技術非常成 熟的標準，其中最常見到的有工s〇（j n t e犷n a t i 〇 η ^丄〇 r 國際組織所财的MPEG系列；另—個則是由！ Tu ( r ^ t匕 …⑽“ Telec〇mm— nica"〇n υη“ )國際組織所做的H . 2 6 x系列，這兩組織後來共同成立一個團體 (Joint Video Team)，共同制定了 η·2。 pEG4 Part 1〇此標準’為目前最新的影像壓縮標準，主 目的是為了加速多媒體通信的發展，並促成電信、_、廣播電視二大 路的融合’祕在壓縮品質或是壓縮效果上，都有非常亮眼的表現^ 網路上的電爾缝灿pEG — 4的梅義，如第三代〜 200922334 通訊中的視訊服務也是。 數位視訊_技術發展已超過十年，#然也有小部份的演變，由不同 的研究理論根據大致可以歸納為下列幾種： 第-種為以模組為基_ ’將影魅面场文字、背景關形視為不 同物件，即物件模組化（Qbject mQdeling)，並制定以物 件為基礎財絲壓·像晝面，鱗到最有鱗的壓财式。在此一概 念中’將動態影像序列中的每-個場景，看成由不同的視訊物件平面所組 成。而同-物件的連續視訊物平面稱為視訊物件，其視訊物可以是動態影 像序列中的人或物或背景，即所輸人_態影像，通過分析可將其分割成 數個視訊物件，騎個視訊物件再對其中所組成的視訊物件分別平面進行 編碼、儲存和傳送。至於解石馬端再依不同的應用來重組、刪減或是替換所 需的視訊物件平面。 第二種方式為_過程沒有做物件模組化，即壓縮方式砂—雜影 像畫面為主題進行的’將每張輸人_像皆視為視訊訊號（S i运^ b a s e d ) ’將影像峨㈣餘資料矛酬除，耻，整張晝面上的所 有圖形、物件及背景’都制相同的壓縮編碼方式，軸產生出來的資料 量相對的也較大，但她於物件模減的高複雜度，且效率低，視訊訊號 擁有較簡單的雜及合理的編碼增益，所以物件餘化編碼方式大多應用 在影像合成或處理上。 為達到上述數位視訊壓職術，可透過使用移動補償系統和使用移動 補償渡波ϋΙ；絲完成，而飾纖純又分為鱗散餘弦縣礎和以離 6 200922334 散小波為基礎，·移動補償濾波器系統分為以時間域和頻率域移動補償編碼 技術’其中，又以小波轉換的延展性（scaiability)被廣範 的討論’基於小波影像的編解碼有2種做法，在頻域（或稱為小波域）上 (1 n —b a n d)和空間域上（s p a t i a 1 )。 而頻域執行編解碼的優點為： 1)編解碼端更容易實現空間的延展性（spatial seal ability)。 2)在較低的f r a m e r a t e下有較好的效能。 3 )在解析雜差的情況下有較好的效能。 然，影像_離散小波轉換，再執行飾_，由於離散小波轉換 有移動變娜（Shift variant川^…必 須利用屬於移動不變異的補償離散小波轉換（_τ)來_ ，而傳統的低頻帶相位平移法所需 移法來估算爾齡纖縣相位平 到移動估_運算複雜度。 輪存㈣，記憶體太大間接影響 是以，要如何解決上述習用 此行業之相關廠商所亟欲研究改 之問題與缺失，即㈣讀明人與從事 善之方向所在者。 200922334 【發明内容】 故’發明人有鑑於上述缺失，乃搜集_資料，經由多方評 ，並以從事於此耔晋 仃業累積之夕年經驗，經由不斷試作及修改，始設 飾用離放小波轉換和移動估測方絲達到顧效果，而達到運算複雜戶 和所需記憶體均較傳統方法來的少之適用於小波視訊編碼之頻又 移動估測方法的發明專利者。 X職 本發明之主要目的在於：先將影像經過_小波轉換，再執行移動估 測’·但離散小波轉換有移動變異特性（s h i f t v a r . +200922334 IX. Description of the Invention: [Technical Field] The present invention provides a wavelet video coding method, which uses discrete wavelet transform and motion estimation methods to achieve compression effects, and achieves computational complexity and required memory compared with conventional methods. Less of the low complexity frequency domain motion estimation method for wavelet video coding. [Prior Art] According to the rapid growth of information, the amount of audio and video data is getting larger and larger, and the field of application has also undergone tremendous changes. The importance of turning to Wei is also thief coming and going, and the Internet is booming. Under, Xu? The capital has been reduced and the heat has been discussed and researched. After the development of the eve of the year, the technology of the hall has also been significantly improved. The state has a very mature technology, and the most common ones are s 〇 (jnte犷nati 〇η ^丄〇r MPEG series of international organizations; another one is by! Tu ( r ^ t匕...(10) “ Telec〇mm— nica"〇n υη”) The H. 2 6 x series, the two organizations later formed a group (Joint Video Team) to jointly develop η·2. pEG4 Part 1 〇 This standard is the latest image compression standard, the main purpose is to accelerate multimedia The development of communication, and promote the integration of telecommunications, _, radio and television two roads 'secret in compression quality or compression effect, have a very bright performance ^ on the network of the electric sewing can pEG-4 of the Yiyi, such as The third generation ~ 200922334 The video service in the communication is also. Digital video technology has been developed for more than ten years, but there are also a small part of the evolution. The different research theories can be summarized into the following categories: Module based _ ' Think of the shadow field and the background as different objects, that is, the object is modularized (Qbject mQdeling), and the object is based on the financial pressure, like the face, the scale to the most scaly pressure type. In this concept, each scene in a sequence of motion pictures is considered to be composed of different video object planes. The continuous video object plane of the same object is called a video object, and the video object can be in a motion image sequence. The person or object or background, that is, the input image of the human being, can be divided into several video objects by analysis, and the video object can be encoded, stored and transmitted separately by riding a video object. Ma Duan then reorganizes, subtracts or replaces the required video object plane according to different applications. The second way is that the _ process does not make the object modularization, that is, the compression method sand-hybrid image is the subject of the theme Each input _ image is regarded as a video signal (S i Yun ^ based ) 'The image 峨 (4) 资料 资料 矛 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , the way The amount of data generated by the axis is relatively large, but she has a high complexity and low efficiency in the reduction of the object. The video signal has a simpler and reasonable coding gain, so the object remnant coding method is mostly applied to the image. Synthetic or processing. In order to achieve the above-mentioned digital video press, it can be done by using the motion compensation system and using the mobile compensation wave; the silk is finished, and the fiber is purely divided into the scales of the cosine county and the scattered waves from 6 200922334 Basic, · Motion compensation filter system is divided into time domain and frequency domain motion compensation coding technology', among which the scaiability of wavelet transform is widely discussed. 'The wavelet image based codec has two methods. In the frequency domain (or called the wavelet domain) (1 n - band) and the spatial domain (spatia 1). The advantages of performing codec in the frequency domain are as follows: 1) The codec side is more likely to implement spatial sealing ability. 2) It has better performance at lower f r a m e r a t e. 3) It has better performance in the case of analyzing the noise. However, the image_discrete wavelet transform, and then perform the decoration _, because the discrete wavelet transform has a moving change (Shift variant Chuan... must use the compensated discrete wavelet transform (_τ) belonging to the mobile non-variant _, and the traditional low-band phase The translation method needs to be used to estimate the phase-to-mobile estimation of the age-old fiber county. The memory complexity (4), the memory is too large, the indirect effect is how to solve the above-mentioned practices of the relevant manufacturers in this industry. The problems and lacks, that is, (4) the person who reads the direction of the person and the direction of the good. 200922334 [Inventive content] Therefore, the inventor has collected the _ information, and has collected the _ The experience of the eve of the year, through continuous trials and modifications, began to set up the use of off-wavelet conversion and mobile estimation of the square wire to achieve the effect, while the realization of complex computing and the required memory are less than the traditional method for wavelet The invention patent of the video coding frequency and mobile estimation method. The main purpose of the invention is to first convert the image through _wavelet conversion and then perform motion estimation. · However, discrete wavelet transform has moving variability (s h i f t v a r . +
Vdriant p pe r t y) ’必須利用屬於移動不變異的補償離散小波轉換（ 丁）來執行移動估測，而低頻帶相位平移法（i〇w “nd su f t)為補償離散小波轉換的具體實現；但傳統的低頻帶相位平移法所需 記憶體太大，以執行3階低頻帶相位平移法來估算，則需要原來工〇倍記 憶體來館存資料，記憶體太大間接影響到移動估測的運算複雜度。本發明 簡化讎她恤，娜1 相位平移法，在較高的階層則使用離散小波轉換，以3階來估算，所需記 憶體為原來的4倍·，再糊離散小轉換之财同階層之子母層關係和移 動向里（motion vector)共用觀念進行移動估測，藉此移 動估測可降低9 0%的運算複雜度。 200922334 【實施方式】 為達成上述目的及功效，本發明所採用之技術手段及構造，茲繪圖就 本發明較佳實施例詳加說明其特徵與功能如下，俾利完全了解。 請參閱第一圖所示，係為本發明較佳實施例之方塊圖，由圖中可清楚 看出本發明係包括： -部份低頻帶相位平移，將-參相像執行部份低鮮相位平 移； -移動估測模組C ’接收該部份倾帶相位平移模㈣傳輪之參考圖 像’並進行移動估測； -移動補償模組D ’接收該移動估測模虹傳輸之參考，並進行 移動補償； -離散小雜換模組Β ’料部之—輪人影似進行離散小波轉換， 產生-目前圖像’該目觸像與經該移動補償敝D處理過之參考圖像執 行相減運算並產生一殘餘訊號； 一編碼模組G ’將該殘餘峨轉為位元流Η。 明同時參閱第-、二圖所示’係為本發明較佳實施例之方塊圖與流程 圖由圖中可，月楚看出，藉由上述構件動作時，依照下列步驟進行·· (10)開始； (1 1)目則圖像執行離散小轉換（discrete wavele t transform，DWT); (1 2 )參相像執行部份低解她平移（P L S B ); 200922334 13)部份低頻帶相位平移係數内差； (14) 執行移動估測（mot i 〇n (15) 執行移動補償(m〇t i on (16) 執行相減運算； e s t i m a t i 〇 n }. compensa t i 〇n)； (1 7)執行編;。 於上述步麵行時，輸人騎錄_散小波職模㈣產生目前圖 像目月ij圖像依序執行離散小波轉換（discrete Wavele t ^ r a n s f 0 rm，DWT)以及由部份低頻帶相位平移模組e進 行部份低頻帶相位平移（p L s B )，將部份低頻帶相位平移係數内差由移 動估測模組C執行移動估測，續以移動補償模組D執行移動補償，且執行 相減運算產生殘餘訊號，將殘餘訊號輯碼模組G執行編碼產生輸出位元 流。 請參閲第三圖所示，係為本發明之部份低頻帶相位平移法示意圖，由 圖中可清楚看出’將原始影像往右（X軸方向）、往下（y轴方向）和往右 再往下（往X軸再往y軸方向）移動，此時將有3張影像，再加上原始影 像，-共有4張影像，將這4張影像進行離散小波轉換（d i s c r e t e wavelet transf〇rm，DWT)，將產生4個ll、 4個HL、4個LH和4個HH sub~band ;圖中的HL (〇， 〇)、LH (〇 ’ 〇)、HH (〇 ’ 〇)和LL (〇，0)表示為原始影像 經過 DWT 產生 4 個 sub-band;HL(l，〇)、lh(1，〇) 、HH (1，〇)和LL ( 1 ’ 〇)表示為往右（叉軸）方向移動後的影 200922334 =行D:的4個“;接著再對4個 此。su — an綠行DWT，而沒有移動的動作，第3階也是如 疋以’本發明之適用於小波視訊編碼之低 可改善制之技術_在於： 賊娜估測方法為 動==雜繼侧換，賊綱帅吻侦轉換有移 動變異特性（shlft variant p p仁r t y )，必須利 、麵（卿τ)錢行移動師' =頻帶相位平移法（1時band shift)為補伽散小波轉 的具_;藉_咖帶她平縣，只«!階執行補償離散小 波轉換也就是低解相位恤，在較高的_使轉散小波轉換，以 3階來估算，所需記憶體為原來的4倍；再利用離散小波轉換之間不同階 層之子母層關係和移動向量（mQtiQn yectQr)共用觀念進 订移動估測’藉此移動估測可降低9 〇%的運算複雜度。 :’以上所述僅為本發明之較佳實施例而已，翻此即拘限本發明之 專利範圍’故舉凡本酬說轉及圖朗容所為之簡綠飾及等效結 構篆化’均朗理包含於本發社專纖圍内，合予陳明。 紅上所述’本發明之適用於小波視訊編碼之低複雜度頻域移動估測方 备於使用時树實能相其功效及目的，故本發明誠為—實雜優異之 發明’為符合發明專利之申請要件，爰依法提出申請，盼審委早日賜准 案乂保障發明人之辛苦發明’倘若鈞局審委有任何稽疑，請不吝來 11 200922334 函指示，發明人定當竭力配合，實感公便。 12 200922334 【圖式簡單說明】 第一圖係為本發明較佳實施例之方塊圖。 第二圖係為本發明較佳實施例之流程圖。 第三圖係為本發明之部份低頻帶相位平移法示意圖。 【主要元件符號說明】Vdriant p pe rty) 'The mobile discrete estimation must be performed using the compensated discrete wavelet transform (D) belonging to the mobile non-variant, and the low-band phase shift method (i〇w “nd su ft) is the specific implementation of the compensated discrete wavelet transform; However, the traditional low-band phase shifting method requires too much memory to perform the 3rd-order low-band phase shifting method to estimate the original memory, which requires the original memory to store the data. The memory is too large and indirectly affects the motion estimation. The computational complexity of the invention. The invention simplifies her shirt, Na 1 phase shifting method, uses discrete wavelet transform in the higher level, estimates in 3 steps, and the required memory is 4 times of the original. The conversion of the same level of the parent-child relationship and the movement vector (motion vector) sharing concept for mobile estimation, thereby reducing the computational complexity by 90%. 200922334 [Embodiment] To achieve the above objectives and effects The technical means and structure adopted by the present invention are described in detail in the preferred embodiments of the present invention. The features and functions are as follows. For a block diagram of a preferred embodiment of the present invention, it will be apparent from the drawings that the present invention includes: - partial low-band phase shifting, and --parameter image performing partial low-fresh phase shifting; - motion estimation module C 'Receiving the reference image of the partial tilt phase shifting mode (four) transmitting wheel' and performing motion estimation; - the mobile compensation module D' receives the reference of the motion estimation mode transmission and performs motion compensation; - discrete small Miscellaneous module Β 'Material part - wheel phantom seems to perform discrete wavelet transform, generate - current image 'the target image and the reference image processed by the motion compensation 敝 D perform subtraction operation and generate a residual signal A coding module G' converts the residual 峨 into a bit stream. See also the first and second figures are a block diagram and a flow chart of a preferred embodiment of the present invention. It can be seen that when the above components are operated, the following steps are performed: (10) starts; (1 1) the image is subjected to discrete wavelet transform (DWT); (1 2) the reference image execution unit Low solution her translation (PLSB); 200922334 13) part of the low frequency band (1) Performing a motion estimation (mot i 〇n (15) performing motion compensation (m〇ti on (16) performing subtraction; estimati 〇n }. compensa ti 〇n); (1 7) Execution editing; In the above steps, the input _ _ wavelet function (4) produces the current image ij image sequentially performs discrete wavelet transform (discrete Wavele t ^ ransf 0 rm, DWT) and Part of the low-band phase shifting module e performs partial low-band phase shifting (p L s B ), and the difference of the low-band phase-shifting coefficient is performed by the motion estimation module C, and the motion compensation is continued. The module D performs motion compensation, and performs a subtraction operation to generate a residual signal, and performs residual encoding of the residual signal modulo module G to generate an output bit stream. Please refer to the third figure, which is a schematic diagram of a part of the low-band phase shifting method of the present invention. It can be clearly seen from the figure that 'the original image is turned to the right (X-axis direction), downward (Y-axis direction), and Move to the right and down (to the X axis and then to the y axis), there will be 3 images, plus the original image, - a total of 4 images, the four images will be discrete wavelet transform (discrete wavelet transf 〇rm, DWT), will produce 4 ll, 4 HL, 4 LH and 4 HH sub~band; HL (〇, 〇), LH (〇' 〇), HH (〇' 〇) in the figure And LL (〇, 0) means that the original image is generated by DWT to generate 4 sub-bands; HL(l, 〇), lh(1, 〇), HH (1, 〇), and LL (1 ' 〇) are expressed as Shadows after moving in the right (fork axis) direction 200922334 = 4 lines of row D:; then 4 more. su - an green line DWT, without moving action, the 3rd order is also like 'the invention The technology that can be applied to the low-visibility system of wavelet video coding _ lies in: The thief-nest estimation method is dynamic ==heterogeneous side-changing, and the thief-class handsome kiss detection has mobile variation characteristics (shlft variant pp rt y), must be profit, face (Qing τ) money line mobile division ' = band phase shift method (1 hour band shift) for the gamma scattered wavelet turn _; borrow _ café with her Ping County, only «! order execution Compensating discrete wavelet transform is also a low-resolution phase-shifting shirt. In the higher _, the divergence wavelet transform is performed, and the estimated memory is 4 times. The memory is different from the original hierarchical layer. The relationship and motion vector (mQtiQn yectQr) share the concept of moving the mobile estimate 'by this mobile estimation can reduce the computational complexity of 9 〇 %. 'The above is only a preferred embodiment of the present invention, and this is the case. The scope of the patents of the invention is limited to the fact that the rewards and the equivalent structure of the syllabus and the equivalent structure of the syllabus are included in the special hairline of the hair salon and are given to Chen Ming. The invention is applicable to the low-complexity frequency-domain mobile estimation of wavelet video coding, and the utility model has the advantages and the purpose of using the tree, so the present invention is an invention of the invention. The application requirements, 提出 apply in accordance with the law, and hope that the trial committee will give an early guarantee The inventor's hard work invented 'If there is any doubt in the ruling committee, please don't hesitate to give us the instructions in 200922334. The inventor will try his best to cooperate with him. 12 200922334 [Simplified illustration] The first picture is based on the invention. The second embodiment is a flow chart of a preferred embodiment of the present invention. The third figure is a schematic diagram of a part of the low frequency band phase shifting method of the present invention.
輸入影像··......AInput image··......A
離散小波轉換模組····BDiscrete Wavelet Transform Module····B
移動估測模組......CMobile estimation module...C
移動補償模組......D 部份低頻帶相位平移模組·E 反離散小波轉換模組···F 編碼模組········0 位元流.........Η 13Motion compensation module...D Partial low-band phase shifting module·E Inverse discrete wavelet transform module···F-encoding module·········0 bit stream... .....Η 13