TWI241074B

TWI241074B - Image compression system using two-dimensional discrete wavelet transformation

Info

Publication number: TWI241074B
Application number: TW92130954A
Authority: TW
Inventors: Bing-Fei Wu; Chao-Jung Chen; Jung-Jeng Chiu; Yan-Lin Chen; Chung-Yan Su
Original assignee: Bing-Fei Wu
Priority date: 2003-11-05
Filing date: 2003-11-05
Publication date: 2005-10-01
Also published as: TW200516868A

Abstract

The invention is related to a kind of image compression/decompression structure using two-dimensional discrete wavelet transformation. The invention uses the wavelet transformation such that it is capable of increasing the compression ratio of input image. In addition, by using the zero-tree coding technique, the amount of data processing required for the input image to conduct the encoding/decoding process is reduced such that the compressing speed of input image can be increased. In addition, through the incorporation of animating interpolating compensating compression technique, the amount of transmitted data can be decreased. When executing the quantifying treatment, it is capable of setting the step value to set the compression ratio for the encoding/decoding process. Thus, in addition to having lower distortion of image, both the amount of image data and the operation processing-time can be reduced so as to realize the purpose of real-time transmitting image.

Description

1241074 玖、發明說明：【發明所屬之技術領域】本發明係關於一種影像壓縮之系統，尤指一種使用二維離散小波轉換的影像壓縮之系統。 5 【先前技術】多媒體之應用在目前日常生活中，已逐漸普及並受到重視。而人與人之間之溝通，已不再局限於語音，更要求月匕夠傳遞即時之影像至遠方的另一位使用者。因此，即時 10 影像壓縮及傳送技術（Real_time Vide〇 compressi〇n and transmission techniques )因而成為近年來通信技術中所關注之焦點，而技術上所遇到之最大困難乃在於：如何有效且即時地壓縮影像資料，不但能將影像資料量減少、壓縮所耗時間減短，且影像資料不致於過度失真而難以辨識。 15此技術所應用之領域十分廣泛，例如：遠距離會議、保全監視系統、影像電話等。習知影像壓縮之技術係採用轉換編碼之技術 (Transf0rm coding)，而目前最普遍的轉換編碼技術乃使用是離散餘弦轉換（Discrete C〇sine Transfer，DCT)，此 20技術已成為聯合圖像專業團體（JpEG)所使用之標準。jpEG 的壓縮倍率一般係為2〇倍至4〇倍之間，而其壓縮後之影像會產生方塊效應，並非執行即時影像壓縮之良好解決方案使用離散小波轉換（discrete wavelet transfomi，DWT) 所得到之壓縮影像，不但壓縮效果較〇(：：7為佳，且無方塊 1241074 效應，因此’愈來愈多即時影像壓縮之方法係改採用dwt 之技術，以取得較佳效果，$包括最新版本之jpeg 2〇〇〇。其中，採用DWT技術之JPEG 2_，在影像壓縮技術方面較習知JPEG具有下列優點： -、高壓縮率：IPEG2_壓縮性能tUpEG提高了 3〇% 〜50%，也就是說，在同樣的圖像質量下，ipEG2綱可以使圖像文件的大小比JPEG圖像文件小3〇〜5〇%。同時，使用JPEG2_的系統敎性好，運行平穩，抗干擾性好，易於操作。 10 15 二、可支援有損和無損壓縮：JPEG只支援有損壓縮， = JPEG2G嶋支援無損壓、缩，無損壓縮僅是將資料量壓縮，所有貧料仍S ;而有損I縮係將部份資料+以省略，以達到更佳之資料壓縮之效果。二、渐進傳輸：傳送圖像資料時，可以先傳輪圖像的輪廊，然後逐步傳輸資料，不斷提高圖像之質量，讓圖像由朦臟到清晰顯示。四可由使用者指定圖像中之爻援感興趣區域部份區域及壓縮效果。然而’當影像資料量很大時，例如，影像資料係 =台監視器之即時影像，而非單觸—台監視器之即時〜目前DWT技術並無法提供足夠快速之影像壓縮速^或可谷§午之失真度’因而造成即時影像產生㈣、失真、或跳格等現象’這並非是使用者所預見的。 20 1241074 【發明内容】之主要目的係在提供—種使用二維離散小波換^像昼縮之系統，俾能提供較佳之影像I缩速度。轉換的Hrr 一/的係在提供一種使用二維離散小波、、和堅縮m ’俾能提供可調整之影像麼縮比。 #用上述目的’本發明揭露一種影像遂縮之系統， =輸即時影像’其包括：順向色彩元素轉換模組，換=輸人輸人影像並進行色彩元素轉換，以輸出色彩轉 10 15 ^像’緩衝模組’係用以輸入色彩轉換影像，並輸出之月】色办轉換影像，動晝差值谓測模組，係用以輸入色彩轉換影像以及之前色彩轉換影像’取其差值並輸出之；順向小波轉換模組，係用以輸入色彩轉換影像及差值，並對其進行小波轉換，以產生小波訊號並輸出之；量化模組，係用以輸入小波訊號’並依據步階值以對其進行純量量化之處理’以產生置化係數並輸出之；以及低記憶體低計算量零树編碼換組，係用以輸入量化係數，並對其進行零樹編碼處理，以產生編碼影像。【實施方式】如圖1所不，本發明二維離散小波轉換的影像壓縮之系統10係包括下列模組：順向色彩元素轉換模組U、動晝差值偵測杈組12、緩衝模組丨3、順向小波轉換模組丨4、量化权組15、及低記憶體低計算量零樹編碼模組16 (Low-Complexity and Low-Memory Entropy 12410741241074 (1) Description of the invention: [Technical field to which the invention belongs] The present invention relates to a system for image compression, and more particularly to a system for image compression using two-dimensional discrete wavelet transform. 5 [Previous Technology] The application of multimedia has been gradually popularized and valued in daily life. The communication between people is no longer limited to voice, but also requires the moon dagger to be able to deliver real-time images to another user in the distance. Therefore, real-time video compression and transmission techniques (Real_time Video Compression and transmission techniques) have become the focus of attention in communication technology in recent years, and the biggest difficulty encountered in technology is how to compress efficiently and in real time The image data can not only reduce the amount of image data, shorten the time taken for compression, but also prevent the image data from being excessively distorted and difficult to identify. 15 This technology has a wide range of applications, such as: long-distance conferences, security surveillance systems, video telephony, etc. The conventional image compression technology uses Transf0rm coding. The most common conversion coding technology currently used is Discrete Cosine Transfer (DCT). These 20 technologies have become the joint image profession. Standards used by the JpEG. The compression ratio of jpEG is generally between 20 and 40 times, and the compressed image will have a block effect. It is not a good solution for performing real-time image compression. It is obtained by using discrete wavelet transfomi (DWT) The compressed image has better compression effect than 0 (:: 7 and no block 1241074 effect, so 'more and more real-time image compression methods are changed to use dwt technology to obtain better results, including the latest version Jpeg 2000. Among them, JPEG 2_ using DWT technology has the following advantages over conventional JPEG in terms of image compression technology:-High compression rate: IPEG2_ Compression performance tUpEG is improved by 30% ~ 50%, also That is to say, under the same image quality, the ipEG2 program can make the size of the image file 30 to 50% smaller than the JPEG image file. At the same time, the system using JPEG2_ has good performance, stable operation, and anti-interference Good, easy to operate. 10 15 2. Support lossy and lossless compression: JPEG only supports lossy compression, = JPEG2G2 Supports lossless compression and compression. Lossless compression is only compressing the amount of data. Still S; and the lossy I shrink is to omit part of the data + to achieve better data compression effect. 2. Progressive transmission: When transmitting image data, you can first transfer the image gallery, and then gradually Transfer data to continuously improve the quality of the image, so that the image changes from dim to clear. Fourth, the user can specify the area of interest in the image and the compression effect. However, when the amount of image data is large For example, the image data is the real-time image of the monitor, not the one-touch—the real-time of the monitor ~ The current DWT technology cannot provide a fast enough image compression speed ^ or the distortion of noon ', which results in real-time The phenomenon of image distortion, distortion, or grid skipping is not expected by the user. 20 1241074 [Summary of the Invention] The main purpose is to provide a system that uses two-dimensional discrete wavelet to transform image like daylight shrinkage. Provides a better image reduction speed. The converted Hrr // system is to provide a two-dimensional discrete wavelet, and a compact m 'm can provide an adjustable image reduction ratio. #Use the above purpose' the present invention Reveal a system of image shrinking, = input real-time image ', which includes: forward color element conversion module, change = input input image and perform color element conversion to output color to 10 15 ^ like' buffer module ' It is used to input the color-converted image and output the month.] The color office converts the image, and the day-to-day difference measurement module is used to input the color-converted image and the previous color-converted image. Take the difference and output it; forward Wavelet conversion module is used to input color conversion image and difference value and wavelet transform it to generate wavelet signal and output it. Quantization module is used to input wavelet signal and use it to step it. Perform scalar quantization processing 'to generate and output the coefficients; and low-memory and low-computation zero-tree encoding swapping, which is used to input quantization coefficients and perform zero-tree encoding processing on them to generate encoded images. [Embodiment] As shown in FIG. 1, the two-dimensional discrete wavelet-transformed image compression system 10 of the present invention includes the following modules: forward color element conversion module U, dynamic day-difference detection branch group 12, buffer mode Group 丨 3, forward wavelet transform module 丨 4, quantization weight group 15, and low-memory low-computation zero-tree coding module 16 (Low-Complexity and Low-Memory Entropy 1241074

Coder，LLZC enc〇der)。於下述中，將對本發明二維離散小波轉換的影像壓縮之系統1〇之運作，作詳細之說明。順向色形元素轉換扠組1丨，係用以讀取輸入影像s〇，並輸出色彩轉換影像S卜其功用乃在於移除輸入影像s〇中 5影像色彩元素間的相關性，以取到較佳之壓縮效果。在此係採用可逆色彩轉換（Reversible c〇1〇r 丁⑽^随，rct)處理，並依下列方程式以取得下列色彩平面： -Y〇 =Coder, LLZC encoder). In the following, the operation of the two-dimensional discrete wavelet-transformed image compression system 10 of the present invention will be described in detail. The forward color shape element conversion fork group 1 丨 is used to read the input image s〇 and output the color conversion image S. Its function is to remove the correlation between the 5 image color elements in the input image s〇 in order to obtain To better compression effect. In this system, we adopt the reversible color conversion (Reversible c0101r Ding ⑽, rct) processing, and obtain the following color plane according to the following equation: -Y〇 =

10 y2=r - G 另外，為了提其中，運算子[]係表示取最大整數值d ^ 高影像資料處理的速度，順向色彩元素轉換模組u執行 rct轉換時，同時對色彩平面進行取樣處理。其中，取樣處理之方式有許多，例如使用4:丨：丨之取樣處理，如圖二 15所示，將方程式⑴、⑶所取得之γι、γ2色彩平面的每四個點的值Γ竇心、Et本一、A τ ^值I貝^點表不）作平均，以取得一個取樣值 r 丁^使Y0 · Y1 · Y2色彩平面之資料量的比例變為4 : 1 : 1。曰^於自然界之影像訊號’其特色為低頻帶聚集較多能 20量，南頻帶所，集之能量較少，因此，可針對色彩轉換影像S 1之低頻及:頻f作進一步的解析，以取得色彩轉換影像S1之低v員及阿頻訊5虎。於本發明中，將順向色彩元素轉換模組n所輸出之色彩轉換影像幻，輸入至順向小波轉換核組14 (Forward DWT)以進行DWT處理，以產生小波訊 1241074 戒S2。順向小波轉換模組i4可將色彩轉換影像^之影像分解成兩個頻帶，如下列方程式（4)及方程式（5)所示： ♦] = 2>[坳0[2«-灸] k (4) k (5) 其中，心/W表示色彩轉換影像S1，N為色彩轉換影像si的長度；心仏/、表示…經過DWT處理後所付之低頻及咼頻訊號；表示DWT處理時，所使用的低通及咼通渡波器。當然，為了取得更精簡之低頻訊號，亦可將…M7再經過一次DWT處理，以取得低頻帶 10中更細緻的小頻寬低頻訊號，即多重解析（multiresolution) 的觀念。例如，本發明係採用三重DWT處理以取得較精簡之低頻及高頻訊號，但並不以此為限。如圖3所示，此為原始之圖播（或為輸入影像S0之其中一幅影像），圖4表示經三重DWT處理處理後，所取得之圖檔。 15 為了提高DWT處理之執行速度，較佳係採用二維Haar D WT，並藉由執行一次水平方向之一維Haar D WT，再執行一次垂直方向之一維Haar DWT，以達到相同之目的，但並不以此為限。其中，一維Haar DWT之低頻係數係如下述方程式（6 )所述，高頻係數係如下述方程式（7 )所述：10 y2 = r-G In addition, in order to mention that the operator [] means to take the maximum integer value d ^ to increase the speed of image data processing, the forward color element conversion module u performs the rct conversion while sampling the color plane. deal with. Among them, there are many ways of sampling processing. For example, using the sampling processing of 4: 丨: 丨, as shown in Fig. 15, the value of each four points of the γι and γ2 color planes obtained by equations ⑴ and ⑶ is Γ , Et, I, A τ ^ value (I ^ ^ points) are averaged to obtain a sampling value r ding ^ so that the ratio of the data amount of the Y0 · Y1 · Y2 color plane becomes 4: 1: 1. The image signal 'in the natural world' is characterized by more energy in the low frequency band and more energy in the lower frequency band. Therefore, the low frequency and frequency f of the color conversion image S 1 can be further analyzed. In order to obtain the low-V member of the color-converted image S1 and Apinx 5 Tiger. In the present invention, the color conversion image magic output from the forward color element conversion module n is input to the forward wavelet conversion kernel group 14 (Forward DWT) for DWT processing to generate wavelet news 1241074 or S2. The forward wavelet conversion module i4 can decompose the image of the color converted image ^ into two frequency bands, as shown in the following equations (4) and (5): ♦] = 2 > [坳 0 [2 «-moxibustion] k (4) k (5) where heart / W indicates the color conversion image S1, and N is the length of the color conversion image si; heart // indicates the low-frequency and high-frequency signals paid after DWT processing; it indicates the DWT processing The low-pass and high-pass ferrule used. Of course, in order to obtain a more streamlined low-frequency signal, the M7 can also be subjected to a DWT process to obtain a more detailed low-band signal in the low frequency band 10, which is the concept of multiresolution. For example, the present invention uses triple DWT processing to obtain more compact low frequency and high frequency signals, but it is not limited to this. As shown in Figure 3, this is the original picture broadcast (or one of the input images S0), and Figure 4 shows the picture file obtained after triple DWT processing. 15 In order to improve the execution speed of DWT processing, it is better to use two-dimensional Haar D WT, and then execute one-dimensional Haar D WT in the horizontal direction and one-dimensional Haar DWT in the vertical direction to achieve the same purpose. But it is not limited to this. Among them, the low-frequency coefficient of the one-dimensional Haar DWT is as described in the following equation (6), and the high-frequency coefficient is as described in the following equation (7):

[2 π] Η—j=-ciq[2h +1] V2 d{[n] = (6) ^α〇[2η]~ΊΊ d〇[2n + \][2 π] Η—j = -ciq [2h +1] V2 d {[n] = (6) ^ α〇 [2η] ~ ΊΊ d〇 [2n + \]

20 1241074 進行1 於方程式（6)係將時間上連續之色彩轉換影像si sm /、程式⑺係將時間上連續之色彩轉換影像組14 ’亚將結果^7及’輸出至順向小波轉換模 5 10 15 〜執行上述三重DWT處理時，為了減少上述運算式子中 =行乘以4的運算所花f的時間，可於執行上述運算、，於水平轉換與垂直轉換皆完成後，，才將資料作右動作，即執行除以2之處理，以節省上述運算:花另外，將連續之輸入影像S0分解成一幅幅的 ::色彩元素轉換模組U所輸出），則畫面與畫面-：連性將是減低編碼運算能量與儲存空間大小此，將時間上連續之前後二個晝面取差值（由動 =組⑽執行）以取得一個留數晝面，其資訊 =體移動量之多S，並將留數晝面輸^順向小波轉; =L4’J:進行處理。如果留數晝面的能量較大= 、里面又動亦較多’其資訊含量亦相對較多，此時，順 °小波轉換模組14就必須以較多的位元數以進行編碼，反之亦然。 ^ 斤示本發明使用二維離散小波轉換的影像壓统：將連續之影像訊號，分成以1〇幅晝面為-組的 Μ处理耘序’亚將這10幅畫面分成I晝面（起始畫面）、旦面（錢晝面）、&Β晝面（内插晝面）。其中，I晝 J、且旦面編碼的參考，由於其重要性較高，所以於編 20 1241074 5 10 15 =處理時’分配較多的位元數以降低重建影像時所產生，▲”現象，P晝面係由動晝差值偵測模組12所產生，藉由儲存上-幅畫面於緩衝模組13，動畫差值谓測模租^可以偵測目前晝面與與上一幅晝面之差異，並產生？畫面， f由順向小波轉換模組14對P畫面進行編碼。於編碼過程中太將B晝面之資料予以保留，#解碼過程中，再由p畫面以生（使用内插法）B畫面。本發明提供二種資料模式：：般模式及高速模式，而兩者較大之差別乃在於B ^數量之多S °由於B畫面係由前後—幅影像藉由内插 2所計算出之影像，所以其失真度較高，這表示高速模式畫面係較一般模式晝面失真。量化模組15:係用以輸入小波訊號S4，並對小波訊號續订純m (scalarquantization)之處理以輸出量化係數S4。其中，量化模組15具有一個無作用區域 (dead-zone )，可用以消除非預期之訊號。控制量化模组 15運作參數乃由步階值S2 (step size)所設定，而步階值 S2的大小可由使用者設定’如果其值較大則產生較小之量化係數S4,因而使量化係數以更為失真，如此—來，即增加重建輸出影像S10的失真度。步階值32愈大，重建後的輸出影像S10之品質愈差’但是，編碼處理後之編碼影像 s 5的資料量將會變小’ 編碼/解碼所f的時間會縮短。因為量化後的係數’是以無失真的壓縮方式進行量化處理，因此，經解量化處理後可得到各個量化後的係數值。 20 1241074 低記憶體低計算量零樹編碼模組丨6:係用以輸入量化係數S4並輸出編碼影像“，其中，係使用〇ne…以零樹編碼（zerotree coding)以及Golomb_Rice ( G-R)碼之技術。20 1241074 Performing 1 in equation (6) is to convert time-continuous color conversion images si sm /, the program is to convert time-continuous color conversion image group 14 'sub-output the results ^ 7 and' to the forward wavelet conversion mode 5 10 15 ~ When performing the above-mentioned triple DWT processing, in order to reduce the time spent in the operation of the above expression = line multiplied by 4, f can be performed after the above operation is performed and after the horizontal conversion and vertical conversion are completed. Perform the right action on the data, that is, perform the process of dividing by 2 to save the above operations: In addition, the continuous input image S0 is decomposed into a frame of :: output by the color element conversion module U), then the screen and screen- : Connectivity will reduce the encoding operation energy and the size of the storage space. Take the difference between the two successive day surfaces in time (performed by motion = group ⑽) to obtain a remaining number of day surfaces, whose information = volume of body movement As many as S, and will leave a few days to face the wavelet ^ forward wavelet turn; = L4'J: for processing. If the energy on the diurnal surface is large =, and there is more movement in it, its information content is also relatively large. At this time, the CW wavelet conversion module 14 must use a larger number of bits to encode, and vice versa The same is true. ^ Shows that the present invention uses a two-dimensional discrete wavelet transform image compression system: the continuous image signal is divided into an M processing sequence with 10 diurnal planes as a group, and the 10 frames are divided into 1 diurnal planes (from (Starting screen), Dan Nian (Qian Ri Nian), & Β Day Nian (Interpolated Day Nian). Among them, the reference of I and J, and the surface coding, because of its high importance, was edited 20 1241074 5 10 15 = During processing, 'allocated more bits to reduce the reconstruction of the image, ▲ "phenomenon , P diurnal surface is generated by the dynamic diurnal difference detection module 12, by storing the upper-frame picture in the buffer module 13, the animation difference is called the measurement mode ^ can detect the current diurnal surface and the previous one The difference between the diurnal plane and the? Screen, f is encoded by the forward wavelet transform module 14 for the P picture. During the encoding process, the data of the diurnal plane B is kept too. During the #decoding process, the p picture is used to generate (Using interpolation) B picture. The present invention provides two data modes: normal mode and high-speed mode, and the big difference between the two lies in the amount of B ^ S ° Since the B picture is borrowed from the front and back-images The image calculated by interpolation 2 has a higher distortion, which means that the high-speed mode screen is more distorted than the normal mode. Quantization module 15: It is used to input the wavelet signal S4 and renew the wavelet signal purely. m (scalarquantization) is processed to output the quantization coefficient S4. The quantization module 15 has a dead-zone, which can be used to eliminate unexpected signals. The operating parameters of the control quantization module 15 are set by the step value S2 (step size), and the size of the step value S2 Can be set by the user 'If the value is larger, a smaller quantization coefficient S4 is generated, so that the quantization coefficient is more distorted. In this way, the distortion degree of the reconstructed output image S10 is increased. The larger the step value 32 is, the reconstruction The worse the quality of the output image S10 after the 'however, the amount of data of the encoded image s 5 after the encoding process will become smaller' the time for encoding / decoding will be shortened because the quantized coefficients are compressed without distortion The quantization process is performed in this way, so after the quantization process, each quantized coefficient value can be obtained. 20 1241074 Low-memory, low-computation zero-tree encoding module 丨 6: It is used to input the quantization coefficient S4 and output the encoded image ", Among them, the technology of One ... using zerotree coding and Golomb_Rice (GR) code is used.

One pass零樹編碼係用以擷取量化係數S4中具有意義之係 5數（significant coefficients ); G_R碼係用以對上述具有意義之係數進行編碼，其中，G-R碼係為一種可變長度碼/可變長度整數的格式，可降低解碼過程中的複雜度，其較佳係使用基本序列（fundamental sequence)之模式，以降低編碼過程之計算。 10 如圖6所示，經過三重二維DWT處理的頻帶圖 (frequency-band image)，每個頻帶的編號，以兩個英文子和一個下標值來表示。第一個字母，代表水平方向是局頻（H)還是低頻（L)，第二個字母則代表垂直方向是高頻 (H)^是低頻（L)，而下標則是代表該頻帶是在第幾重 15 $理後所產纟的頻帶。LL3頻帶代表的是最低頻的頻帶，它的係數係表示影像中的直流成份（dc c〇efficient)。相同方向的頻帶將組成樹狀結構，樹根的位置係落在最後一重DWT所處理的頻帶，即HL3、LH3、及腦3。其中，箭頭表不每一棵樹後代子孫的方向。除了節點係落在第一重 20 DWT所處理之頻帶外，每一個節點有四個直接的後代子孫。圖中在HH3和HH2頻帶中的方格顯示出這個關係。低記憶體低計算量零樹編碼模組16之編碼方式，係如下列虛擬編碼所示： (1 ) Use DPCM scheme to code the dc coefficients. 12 1241074 (2) For each tree k5 EncodeTree(k){ G-R—FS(k);One pass zero-tree coding is used to capture the significant coefficients of the quantized coefficient S4; G_R code is used to encode the above-mentioned meaningful coefficients, where the GR code is a variable length code The format of the / variable-length integer can reduce the complexity in the decoding process. It is preferably a mode using a fundamental sequence to reduce the calculation in the encoding process. 10 As shown in Figure 6, the frequency-band image after the three-dimensional two-dimensional DWT processing. The number of each frequency band is represented by two English subscripts and a subscript value. The first letter represents whether the horizontal direction is the local frequency (H) or the low frequency (L), the second letter represents the vertical direction is the high frequency (H) ^ is the low frequency (L), and the subscript indicates that the frequency band is The frequency band of the plutonium produced after the weight of 15 $. The LL3 band represents the lowest frequency band, and its coefficient represents the DC component (dc cefficient) in the image. The frequency bands in the same direction will form a tree structure, and the position of the root of the tree falls in the frequency band processed by the last heavy DWT, namely HL3, LH3, and brain 3. Among them, the arrows do not indicate the direction of the descendants of each tree. Except for nodes that fall into the frequency band handled by the first heavy 20 DWT, each node has four direct descendants. The boxes in the HH3 and HH2 bands show this relationship. The encoding method of the low-memory, low-computation zero-tree encoding module 16 is as shown in the following virtual encoding: (1) Use DPCM scheme to code the dc coefficients. 12 1241074 (2) For each tree k5 EncodeTree (k) { GR-FS (k);

If at least one descendant of k is not zero{ Output 1 in 1 bit; 5 For each sub-tree of k: s, EncodeTree(s); } elseIf at least one descendant of k is not zero {Output 1 in 1 bit; 5 For each sub-tree of k: s, EncodeTree (s);} else

Output 0 in 1 bit; }Output 0 in 1 bit;}

10 其中，DPCM的編碼方式是先將第一個直流成份使用 G-R編碼，而接下來的係數則是以相對的直流成份（目前的直流成份減去前一個直流成份）以進行後續之編碼。其中，如圖7所示，G_R—FS(X)所使用之匕R編碼步驟如下： (1)判斷輸出係數X的類別（dass)。若乂屬於第n類， 15則輸出n+1個位元，其中前面的以固位元為〇，只個位元為1。俊則用1個位元來編碼X 一個位元來表示X的符10 Among them, the DPCM coding method first uses the first DC component to use G-R coding, and the next coefficient is the relative DC component (the current DC component minus the previous DC component) for subsequent coding. Among them, as shown in Fig. 7, the encoding steps of G R-FS (X) are as follows: (1) Determine the type (dass) of the output coefficient X. If 乂 belongs to the nth category, 15 outputs n + 1 bits, of which the previous bit is 0, and only 1 bit is 1. Jun uses one bit to encode X and one bit to represent the sign of X.

(2 )若X屬於第^類，且n〉〇，的絕對值中最後的nq個位元，且用唬。〇表示正數，而1表示負數。 20 二如’广的值為_12,則在第_步時，會輸出〇〇〇〇1 =屬於弟四類)，而在第二步時，會輸出100(12的_ 值為1100，後三碼為100)，並負數）。上個付號位兀1(肩藉由本發明二維離即可得到輸出影像 ^當編碼影像S5傳送至使用者端時放小波轉換的影像解壓縮之系統5〇， 13 25 1241074 S10,由於本發明二維離散小波轉換的影像解壓縮之系統 50之運作方式與本發明〔維離散小波轉換的影像壓縮之系統10，下述中僅略作說明。低圮憶體低計算量零樹解碼模組5 i (LOW_complex办 5 and Low-Memory Entropy decoder ’ LLZC decoder)係用以輸入編碼影像S5 ’並對其進行零樹解碼處理以輸出逆向量化係數S6。低記憶體低計算量零樹解碼模組“之運算方式，係為低記憶體低計算量零樹編碼模組16之運算方式的 ^運算。於G-R_FS(x)的反運算部份’首先需判斷\的類別。 10猎由讀取一個一個的位元，並且計算i以前的〇之數目，即可知道該係數的類別。之後，再用取出卜丨個位元加上2n i，即為該係數的絕對值。最後，再解出符號位元即可得到該係數的值。例如，_12於零樹解碼時，丄以前共計有因此，其類別乃為類別4,再取出3位元以得到1〇〇(即十進 15，之4) ’加上23,故可得到數字12，最後，讀取符號位元知1，因此，該係數值為-12。解量化模組52係用以輸入逆向量化係數S6，經解量化處理後可得到各個量化後的係數值，再將各個係數值乘以步階值S2，即可得到解量化處理後的逆向小波訊號”之 20值，並輸出之。逆向小波轉換模組53係用以輸入逆向小波訊號s7，並進行逆向DWT以取得逆向色彩轉換影像S9，並輪出之。其中，逆向色彩轉換影像S9係依據方程式（8)之計算所取 14 1241074 «〇[«] = 2]A[«-2k]a{[k] + -2k]dx[k] ( 8) 及封〃]係分別表示逆向DWT所使用之低通與高通渡波器。若之前係採用二維Haar DWT，則二維Haar逆向 DWT係依據下列方程式以取得逆向色彩轉換影像§9 : 5 i〇[«] - ^j^(cii[n] + d{[n]) (9) a〇[n +1] = _(βι [„] ^ ^ (10)(2) If X belongs to the ^ class, and n> 0, the last nq bits in the absolute value of, and use bluff. 0 indicates a positive number, and 1 indicates a negative number. 20 Second, if the wide value is _12, in the step _, it will output 〇00〇01 = belong to the fourth class), and in the second step, it will output 100 (the value of _12 is 1100, The last three codes are 100), and negative numbers). The previous payout position 1 (the output image can be obtained by shouldering the two-dimensional separation of the present invention ^ when the encoded image S5 is transmitted to the user side, a wavelet-transformed image decompression system 50, 13 25 1241074 S10. The operation mode of the two-dimensional discrete wavelet-transformed image decompression system 50 and the present invention [the system of the two-dimensional discrete wavelet-transformed image compression system 10 are described only in the following. Low-memory low-computation zero-tree decoding module Group 5 i (LOW_complex office 5 and Low-Memory Entropy decoder 'LLZC decoder) is used to input the encoded image S5' and perform zero-tree decoding on it to output the inverse vectorization coefficient S6. Low-memory low-computation zero-tree decoding module The calculation method of group "is the ^ operation of the calculation method of the low-memory, low-computation zero-tree encoding module 16. In the inverse operation part of G-R_FS (x), you must first determine the category of \. 10 猎由Read the bits one by one, and calculate the number of 0 before i, you can know the type of the coefficient. After that, take out the bits and add 2n i, which is the absolute value of the coefficient. Finally, Sign bit The value of this coefficient can be obtained. For example, when _12 is decoded in the zero tree, there is a total of 因此. Therefore, its category is category 4, and then 3 bits are taken out to obtain 100 (that is, decimal 15, 4). Go to 23, so you can get the number 12, and finally, read the sign bit to know 1, so the coefficient value is -12. The dequantization module 52 is used to input the inverse vectorization coefficient S6, which can be obtained after the dequantization process. The quantized coefficient values are multiplied by the step value S2 to obtain the 20 values of the inverse wavelet signal after dequantization processing and output it. The inverse wavelet transform module 53 is used to input the inverse wavelet The signal s7 is subjected to inverse DWT to obtain the inverse color conversion image S9, and it is rotated out. Among them, the inverse color conversion image S9 is obtained according to the calculation of equation (8) 14 1241074 «〇 [«] = 2] A [« -2k] a {[k] + -2k] dx [k] (8) and seal] are the low-pass and high-pass wave filters used for reverse DWT respectively. If the previous two-dimensional Haar DWT was used, the two-dimensional Haar's inverse DWT is based on the following equation to obtain an inverse color-converted image §9: 5 i〇 [«]-^ j ^ (cii [n] + d {[ n]) (9) a〇 [n +1] = _ (βι [„] ^ ^ (10)

當然，如桌順向小波轉換模組14係使用三重二維!^^^ 處理，則逆向小波轉換模組53則使用三重二維逆向DWT處理。 10 如果逆向色彩轉換影像S9係為I晝面，則輸出I畫面至反向色彩元素轉換模組57 (以產生一幅晝面）及緩衝模組 55 ;如果逆向色彩轉換影像仍係為P畫面，則輸出p畫面至差值補償模組54，差值補償模組54依據動晝差值補償 (Inter-Frame Difference Compensation)技術以產生一幅 15畫面，並輸出至反向色彩元素轉換模組57及雙向内插模組Of course, if the table forward wavelet conversion module 14 uses triple 2D processing! ^^^, the inverse wavelet conversion module 53 uses triple 2D inverse DWT processing. 10 If the reverse color conversion image S9 is I daylight, output I picture to the reverse color element conversion module 57 (to produce a daylight face) and the buffer module 55; if the reverse color conversion image is still P picture , Then output the p picture to the difference compensation module 54. The difference compensation module 54 generates a 15 picture according to the Inter-Frame Difference Compensation technology and outputs it to the inverse color element conversion module. 57 and bi-directional interpolation module

56 ;如果逆向色彩轉換影像S9係為B晝面，則輸出b畫面至緩衝模組5 5 ’雙向内插模組5 6依據差值補償模組5 4及緩衝模組55所輸出之資料，進行雙向内插補償（Bidirecti〇nal Interpolation )處理以產生一幅晝面，並輸出至反向色彩元 20 素轉換模組57。反向色彩元素轉換模組57接收來自逆向小波轉換模組53、差值補償模組54、及雙向内插模組56所輸出之晝面， 15 1241074 並進行逆向RCT處理以取得輸出影像sl〇，並輸出之。其中，逆向RCT處理係依據下列方程式： G-y〇-( 11 ) r = Y2+g ( 12) 5 B = r^G ( 13) 本發明二維離散小波轉換的影像壓縮之系統1〇係使用小波轉換技術，故能提高輸入影像训之壓縮比，並藉由零樹編碼（zerotreecoding)之技術，以減少對輸入影像s〇進行編碼/解碼時所需之資料處理量，故能提高輸入影像s〇 10之壓縮速度，並配合動畫内插（Frame Interpolation)補償壓縮技術，可減少編碼影像85之資料量。另外，執行量化處理時，使用者可設定步階值（step size)以設定編碼時的壓縮比（compression rati〇)。最後，將產生的編碼影像S5 U至运☆而並由返立而進行解碼處理，可想而知地，於解馬處理日守’其執行程序為編碼處理之程行程序相反，僅需作些微地調整。藉由上述技術，本發明二維離散小波轉換 ^影像壓縮之系統1〇能滿足使用者對影像之高壓縮速度、高壓縮比之需求。上述實施例僅係為了方便說明而舉例而已，本發明所張之權利範圍自應以申請專利範圍所述為準，而非僅限方；上述實施例。【圖式簡單說明】 16 1241074 圖1係本發明二維齙塊圖。 S小波軺換的影像壓縮系統之方圖2係色彩平面進行取樣處理之示意圖。 5 10 圖3係進行影像I縮時所使用之標準影像。圖4錢料”二轉散小後所得到之壓縮影像。 ϋ細糸統 «5#不㈣縮模式下的影像組成之。圖6係經過三重二維斷處理之頻帶I。圖7係G-R編碼的類別及成份之對照表。 ^ 8係本發明二維離散小波轉換的影像解壓縮系統之万塊圖。【圖號說明】 10 11 13 15 二維離散小波轉換的影像壓縮系統順向色彩元素轉換模組緩衝模組量化模組 12動畫差值偵測模組 14順向小波轉換模組56; if the reverse color conversion image S9 is a B-day surface, output b picture to the buffer module 5 5 'two-way interpolation module 5 6 according to the data output by the difference compensation module 54 and the buffer module 55, Bidirectional interpolation compensation is performed to generate a daylight surface and output to a reverse color element 20 element conversion module 57. The inverse color element conversion module 57 receives the daylight surface output from the inverse wavelet conversion module 53, the difference compensation module 54, and the bidirectional interpolation module 56, 15 1241074, and performs reverse RCT processing to obtain the output image sl. And output it. Among them, the reverse RCT processing is based on the following equation: Gy〇- (11) r = Y2 + g (12) 5 B = r ^ G (13) The two-dimensional discrete wavelet transform image compression system 10 of the present invention uses wavelet The conversion technology can improve the compression ratio of the input image training, and the zerotree coding technology can reduce the amount of data processing required to encode / decode the input image s0, so it can improve the input image s The compression speed of 〇10, combined with the animation interpolation (Frame Interpolation) compensation compression technology, can reduce the data volume of the encoded image 85. In addition, when performing quantization processing, the user can set a step size to set a compression rati at the time of encoding. Finally, the generated coded image S5 U is shipped to ☆ and decoded by returning. It is conceivable that in the solution of the processing of the horse, the execution process is the opposite of the coding process. Slightly adjusted. With the above-mentioned technology, the two-dimensional discrete wavelet transform of the present invention ^ image compression system 10 can meet users' needs for high compression speed and high compression ratio of images. The above embodiments are merely examples for the convenience of description. The scope of rights claimed in the present invention shall be based on the scope of the patent application, rather than being limited to the above; the above embodiments. [Brief description of the drawings] 16 1241074 FIG. 1 is a two-dimensional block diagram of the present invention. S-wavelet transform image compression system Figure 2 is a schematic diagram of the color plane for sampling processing. 5 10 Figure 3 is a standard image used for image I reduction. Fig. 4 Compressed image obtained after the second small divergence. ΫDetailed system «5 # does not shrink the image composition. Fig. 6 is the band I after the three-dimensional two-dimensional cut processing. Fig. 7 is GR Comparison table of coding categories and components. ^ 8 is a 10,000-block diagram of the two-dimensional discrete wavelet transform image decompression system of the present invention. [Illustration of the drawing number] 10 11 13 15 Two-dimensional discrete wavelet transform image compression system forward color Element conversion module Buffer module Quantization module 12 Animation difference detection module 14 Forward wavelet conversion module

^低記憶體低計算量零樹編碼模組 50二維離散小波轉換的影像解壓縮系統 5 1低記憶體低計算量零樹解碼模組 52解量化模組 53逆向小波轉換模組 54差值補償模組 55緩衝模組 %雙向内插模組 57逆向色彩元素轉換模組 17^ Low-memory low-computing zero-tree encoding module 50 Two-dimensional discrete wavelet transform image decompression system 5 1 Low-memory low-computing zero-tree decoding module 52 dequantization module 53 reverse wavelet transform module 54 difference Compensation module 55 Buffer module% Bidirectional interpolation module 57 Reverse color element conversion module 17

Claims

1241074 拾、申請專利範圍： 1· 一種影像壓縮之系統，係用以傳輸即時影像，其包括· 一順向色彩元素轉換模組，係用以輸入一輸入影像並 5進行一色彩70素轉換，以輸出一色彩轉換影像；一缓衝模組，係用以輸入該色彩轉換影像，之前色彩轉換影像；㈣一動晝差值偵測模組，係用以輸入該色彩轉換影像以及該之前色彩轉換影像，取其差值並輸出之； 0 一順向小波轉換模組，係用以輸入該色彩轉換影像及該差值，亚對其進行一小波轉換，以產生一小波訊號並輸出之；里化模、’且係用以輸入該小波訊號，並依據一步階 15 20 值以對其進行-純量量化之處理，以產生—量化係數並輸出之；以及 ^ 一低圮憶體低計算量零樹編碼模組，係用以輸入該量化係數亚對其進行一零樹編碼處理，以產生一編碼影像。一 2·如申請專利範圍第1項所述之系統，其中，該色彩 70素轉換係為可逆色彩轉換。該之前該小波1241074 The scope of the patent application: 1. An image compression system for transmitting real-time images, including: a forward color element conversion module for inputting an input image and 5 for 70 color conversion, To output a color-converted image; a buffer module to input the color-converted image and the previous color-converted image; 动 a day-to-day difference detection module to input the color-converted image and the previous color-converted image Take the difference value of the image and output it; 0 A forward wavelet conversion module is used to input the color conversion image and the difference value, and perform a wavelet conversion on it to generate a wavelet signal and output it; The model is used to input the wavelet signal and perform a scalar quantization process based on a step 15 20 value to generate a quantization coefficient and output it; and a low calculation volume The zero-tree coding module is used to input the quantized coefficients and perform a zero-tree coding process on them to generate a coded image. -2. The system according to item 1 of the scope of patent application, wherein the color 70 element conversion is a reversible color conversion. Before the wavelet

^ 如申請專利範圍第1項所述之系統，士色形轉換影像係為時間上較早之該色彩轉換, 4.如申請專利範圍第1項所述之系統，；換係為二維Haar小波轉換。 18 1241074 其中，該小波其中，該小波及至少一内插 5.如申請專利範圍第1項所述之系統轉換係為三重小波轉換。 6·如申請專利範圍第1項所述之系統訊號係包括一原始晝面、至少一留數晝面 5 畫面。 “ 7.如申請專利範圍第1項所述之系統，其中，該旦仆杈組更具有一無作用區域，係用以 /里〇 , 于、非預期之小波訊號。^ As for the system described in item 1 of the scope of the patent application, the color conversion image is the color conversion earlier in time, 4. The system described in item 1 of the scope of the patent application, and the system is two-dimensional Haar Wavelet transform. 18 1241074 Among them, the wavelet Among them, the wavelet includes at least one interpolation 5. The system conversion described in the first item of the scope of patent application is triple wavelet conversion. 6. The system as described in item 1 of the scope of patent application. The signal consists of an original daylight surface and at least one daylight image. "7. The system as described in item 1 of the scope of patent application, wherein the server group further has an inactive area, which is used for / unintentional wavelet signals.

10 •如“專利範圍第i項所述之系統，其中，如步階值較大，則該量化係數較小，反之亦然。 Μ 9·如申請專利範圍第i項所述之系統’、，、、其中，該編碼處理係使用一〇ne pass零樹編碼。 10·如申請專利範圍第9項所述之系統，其中，該隨 pass零樹編碼係以G〇i〇mb-Ricy^以進行編碼。 ,11.如申請專利範圍第i項所述之系統，其中，該編碼 15 像可由〜像解壓縮之系統以進行解壓縮處理，以輸出 -輸出影像，其中，該影像解壓縮之系統係包括：10 • The system described in item i of the patent scope, wherein if the step value is larger, the quantization coefficient is smaller, and vice versa. M 9 · The system described in item i of the patent scope ', The coding process uses zero-pass zero-tree coding. 10. The system described in item 9 of the scope of the patent application, wherein the zero-pass coding with pass is G0iMB-Ricy ^ 11. The system as described in item i of the patent application range, wherein the encoded 15 images can be decompressed by a ~ image decompression system to perform an output-output image, wherein the image is decompressed The system includes:

-低e憶體低計算量零樹解碼模組，係用以輸入該編碼影像，並對其進行一零樹解碼處理，以產生一解量化係數並輸出之；一解置化模組，係用以輸入該解量化係數，並依據該步階值以對其進行—解量化之處理，以產生一解小波訊號並輸出之； 19 1241074 逆向小波轉換模組，係用以輸入該解小波訊號，並對其進行逆向小波轉換處理，以產生一解色彩轉換影像並輸出之；一緩衝模組，係用以儲存該解色彩轉換影像，並輸出 5 一之前解色彩轉換影像；、/一差值補償模組，係用以輸入該解色彩轉換影像，並進行動^差值補償處理，以輸出一差值解色彩轉換影像；-A low-e memory low-computation zero-tree decoding module, which is used to input the encoded image and perform a zero-tree decoding process to generate a dequantized coefficient and output it; It is used to input the dequantization coefficient and to perform de-quantization processing based on the step value to generate and output a solution wavelet signal. 19 1241074 Inverse wavelet conversion module is used to input the solution wavelet signal. , And inverse wavelet transform it to generate a de-color-converted image and output it; a buffer module is used to store the de-color-converted image and output 5 before-de-color-converted image; The value compensation module is used for inputting the decolorized converted image and performing dynamic difference compensation processing to output a difference decolorized converted image;

、二雙向内插模組，係用以輸入該之前解色彩轉換影像、及孩差值解色彩轉換影像，並進行雙向内插處理，以輸 10出一内插解色彩轉換影像；以及刖史一逆向色彩元素轉換模組，係、用以_入該解色彩轉換影像、該差值解色彩轉換影像、及㈣插解色彩轉換影像，並對其進行-逆向色彩元素轉換以產生一輸出出之。 15 12.如申請專利範圍第11項所述之系、統’其中，該零樹解碼處理係使用一〇ne pass零樹解碼。, Two-way bi-directional interpolation module, used to input the previously de-color-converted image, and the difference-valued de-color-converted image, and perform two-way interpolation processing to output 10 output one-interpolated de-color-converted image; and history An inverse color element conversion module is used to input the de-color-converted image, the difference de-color-converted image, and interpolate the de-color-converted image, and perform inverse-color element conversion to generate an output. Of it. 15 12. The system and system described in item 11 of the scope of patent application, wherein the zero-tree decoding process uses one pass zero-tree decoding.

13. 如申請專利範圍第12項所述之系統，其中，該〇如 pass零樹編碼係以G〇1〇mb-Ricw·^以進行解碼。 14. 如申請專利範圍第n項所述之系統，其中，該逆向 20小波轉換處理係為二維Haar逆向小波轉換處理。、15.如申請專利範圍第n項所述之系統，其中，該逆向小波轉換處理係為三重逆向小波轉換處理。 16·如申請專利範圍第n項所述之系統，其中，該逆向色彩元素轉換係為逆向可逆色彩轉換。 2013. The system as described in item 12 of the scope of patent application, wherein the zero-pass coding of the zero-pass is decoded with G010mb-Ricw · ^. 14. The system described in item n of the scope of patent application, wherein the inverse 20 wavelet transform process is a two-dimensional Haar inverse wavelet transform process. 15. The system according to item n of the scope of patent application, wherein the inverse wavelet transform process is a triple inverse wavelet transform process. 16. The system according to item n of the scope of patent application, wherein the reverse color element conversion is a reverse reversible color conversion. 20