200929894 九、發明說明： 【發明所屬之技術領域】 本發明與提升視訊晝面率有關，特別是指一種能夠降低 記憶體的使用量，改善網路頻寬利用率之方法。 【先前技術】 由於電腦網路的快速發展與多媒體應用的普及，為了降 低影像所需的記憶體容量與有效地使用網路傳輸頻寬，視訊 壓縮技術便因此產生； 在視訊壓縮技術中，以ITU-T H. 261/3/L與ISO MPEG-1/2/4所發展的壓縮標準已被廣泛的接受並應用在各 個層面，特別是當前發展快速的第三代行動網路通訊服務所 提供的在線觀賞影片與即時視訊影像通話，更進一步的拉近 了人與人間的距離； 然而要達到上述的通訊服務，所需的視訊資料量卻相當 龐大，即使透過壓縮技術去除資料的冗餘性，在有限記憶體 與有限頻寬的限制下，往往無法提供令人滿意的畫面品質。 為了改善在有限記憶體與有限頻寬所造成畫面品質不 佳的問題，使視、訊資料在上述條件下能保有可接受的晝面品 質，有人便提出了晝面捨棄的技術，其技術是藉由捨棄部份 5 200929894 原始晝面來減少晝面數量，藉此來解決有限頻寬的問題，為 …σ被捨棄的旦面’提升視訊畫面率之技術便因運而生； 常用的提升視訊晝㈣之方法，簡述如下： 面υ ‘將前―張晝面複製後作為内插晝面插 幸乂為簡早#速’但在重建視訊後，影像中物體的移動會 較不平順。 Ο Ο 2.晝面平均法：是取前後兩張晝面平職產生内插晝 面’改善了物體移動不平順的但内插晝面卻 糊與殘影’尤其在物體的邊緣上更為嚴重。 ▲ 3·私動補仏晝面内插法：是經過移動估計之後，計算出 較佳且具有意義的移動向量，由此產生内插晝面；: 必須先找職差最小的移麵量，導致計算複雜度較高:因 此較不適合應用在即時視訊傳輪上。 可知目前提升視訊晝面率之方法雖然多，但 點，因此本發明人便針對以往的缺點加以研究，進Μ —種利用灰色多項式内插法提升視訊畫面率之方法。& 【發明内容】 内插法提升視訊晝面率之 面壓縮所需要的記憶體容 本發明提出利用灰色多項式 方法，第一目的在於減少視訊書 6 200929894 量，且能夠同時保有可接受的晝面品質。 第一目的為有效改善有限頻寬的使用率。 為達前述目的，此方法以對視訊畫面降低取樣，以提 縮的性能，減少需要的記憶體容量及改善有限 Ο Ο :見：率’在接收視訊晝面後，再估計重建捨棄的書 面，讓視矾畫面能夠保有一定程度的品質； 此方法包含下列步驟： 取樣步驟：將視訊晝面做時間域的降低取樣· 壓縮步驟：將降低取樣之視訊畫面壓縮轉輸； 解壓縮步驟：將壓縮後的視訊晝面解壓縮；以及 重建步鄉：制灰色乡料_ 捨棄的視訊晝面。 τ輯取樣步驟中 【實施方式】 轉_用灰色乡項式_料升視㈣面率之方 貫施例如第1圖至第4圖所示，其包含下列步驟： ::將視訊晝面做時間域的降低取樣，此實施例 個查面心，將此視訊晝面依時間她_列為2Ν+1 個旦面，亚將偶數畫面Λ2、α4、α6 λ 面的奇數晝面·‘、“·=’保留視訊畫200929894 IX. Description of the Invention: [Technical Field of the Invention] The present invention relates to improving the video coverage rate, and more particularly to a method for reducing the usage of memory and improving the utilization of network bandwidth. [Prior Art] Due to the rapid development of computer networks and the popularity of multimedia applications, video compression technology has been created to reduce the memory capacity required for images and to effectively use the network transmission bandwidth. In video compression technology, The compression standards developed by ITU-T H. 261/3/L and ISO MPEG-1/2/4 have been widely accepted and applied at various levels, especially the current fast-developing third-generation mobile network communication services. Providing online viewing videos and instant video video calls further narrows the distance between people; however, to achieve the above-mentioned communication services, the amount of video data required is quite large, even if the redundancy of data is removed through compression technology. Sex, limited by limited memory and limited bandwidth, often does not provide satisfactory picture quality. In order to improve the picture quality caused by limited memory and limited bandwidth, and to ensure that the visual and video data can maintain acceptable quality under the above conditions, some people have proposed a technology to eliminate the problem. The technology is By reducing some of the original number of 200929894 to reduce the number of defects, to solve the problem of limited bandwidth, the technique of increasing the video frame rate for the sigma of σ is born; the common promotion The method of video (4) is briefly described as follows: Face υ 'Copying the front 昼 昼 后 作为 作为 作为 作为 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制 复制. Ο Ο 2. The method of 昼 平均 : : : : : : : 昼 昼 昼 昼 昼 昼 昼 昼 昼 昼 昼 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 平均 ' ' ' ' ' ' ' ' ' ' ' serious. ▲ 3. Private interpolation surface interpolation method: After the motion estimation, the better and meaningful motion vector is calculated, thereby generating the interpolation surface; This leads to higher computational complexity: it is less suitable for use on instant video transmissions. It can be seen that although there are many methods for improving the video coverage rate, the present inventors have studied the shortcomings of the past, and have developed a method for improving the video picture rate by using gray polynomial interpolation. & [Summary of the Invention] Interpolation method for improving the face compression of video face rate The present invention proposes to utilize the gray polynomial method, the first purpose is to reduce the amount of video book 6 200929894, and at the same time maintain acceptable 昼Surface quality. The first purpose is to effectively improve the utilization of finite bandwidth. In order to achieve the above objectives, this method reduces the sampling of the video picture to reduce the performance, reduce the required memory capacity and improve the limited Ο 见: see: rate 'after receiving the video, then estimate the written copy of the remnant, This method can maintain a certain degree of quality; this method includes the following steps: Sampling step: reduce the sampling of the video field in the time domain. · Compression step: compress and reduce the video image of the reduced sampling; decompression step: compress After the video is decompressed; and the reconstruction step township: the gray township _ abandoned video face. In the τ series sampling step, the following steps are performed: The sampling of the time domain is reduced. In this embodiment, the face of the face is checked, and the video face is listed as 2Ν+1 旦 faces according to time, and the even-numbered frames 亚2, α4, α6 λ face odd-numbered faces·, "·='Retained video
Ami，便剩下N+1個畫面； 7 200929894 壓縮步驟:將降低取樣之視訊晝面墨縮並傳輪，此實施 种如W圖所示，將降低取樣後餘下的Ν+ι個奇數書面壓 鈿，並傳輸至接收位置B ; — ”解壓^步驟：將壓縮後的視訊晝面解麵，此實施例中 女第3圖所不’在該接收位f B將壓縮後的㈣個奇數書面 解壓縮；以及 |Ami, there are N+1 pictures left; 7 200929894 Compression step: the video of the reduced sampling is reduced and the wheel is transferred. This embodiment, as shown in Figure W, will reduce the remaining Ν+ι odd number written after sampling. Pressed and transmitted to the receiving position B; — ”Decompression step: The face of the compressed video is unwrapped. In this embodiment, the female figure 3 does not 'compress the (four) odd numbers in the receiving bit f B Written decompression; and |
Ο 重建步驟：利用灰色多項式内插法估計重建在取樣步驟 中捨棄的視訊晝面，此實施例中如第4圖所示，將㈣個奇 數=面使肢色多項式_法，估計出二相鄰奇數晝面之間 的晝面A2、A4、Ae...A2N，讓視訊晝面恢復至2N+1個晝面，其 中灰色多項式内插法的技術敘述如下： ()、‘隹夕項式内插法（polynomial inteipolation, PI) 給定原始育料χ(々)，一維多項式内插法實行步驟如下： 步驟1.令x(幻具有L階多項式的型式，也就是說 x(k) = cLkL + clJcla + …+ Clk + Cq ⑴ 步驟2.將+ 1代入（丨）式，可得 x = Vc (2) 其中X，C與F分別代表x(免)，4與= 0-y For 0印z，+ 1。 步驟3.求解内插值+ i/w) ^k + VM) = cL(k + l/M)L +--- + Cl(k + l/M) + c0(3) 8 200929894 其中係數心為由式（2 )求得。 步驟4.獲得最終的内插值，如下 x(k + 1/M) = x(k + l/M) χ Μ ( 4 ) 其中#表示一個放大係數且一般而言為一個 整數。 在一維多項式内插法裡資料間的隨機性是影響其性能 的主要原因之一。因此，若能降低資料的隨機性，其性能應 ® 可有效的得到改善。 (2)灰色多項式内插法 灰色系統是由鄧聚龍教授於1982年所提出的理論，其 中又以灰色模型GM(1，1)為基礎的應用較為廣泛。在GM(1, 1) 模型中使用灰色生成的方法來降低原始資料數列間的隨機 性。而常見的灰生成方法為一次累加生成（first-order ◎ accumulated generating operation，1-AG0)與一次反累力口生成 (first-order inverse accumulated generating operation， 1-IAGO) ’相關描述如下。 給疋原始資料2 0, K h尺}，根據一次累加生成產 生新的資料χ(1)㈨如下： x{l\k) = Y^x{n) (5) /7=1 for KB[’其中χ(ι)⑴⑴。從式（5)中，可以輕 9 200929894 易地從X⑴(/c)裡找回原始資料 w χ(^) = χ^(^)-χ(1)(^-1) ( 6) for [。這個運算稱為一次反累加生成。 在這裡，我們提出一個灰色多項式内插法（grey polynomial interpolation, GPI)來改善多項式内插法的性能。 由於一次累加生成可有效降低資料間的隨機性，因此在所提 的方法中，我們使用一次累加生成的技巧對原始資料進行前 〇 處理，藉此降低資料的隨機性。接著使用一維多項式内插法 求得内插值。再藉由一次反累加生成後處理獲得内插值。最 後，再使用一個α濾波器對内插值做修正。記MXM分別為水 平與垂直方向放大倍數為Μ倍，已知大小為lxl且色彩格式 為RGB二原色所組成的原始影像，經色彩轉換為YCbCr格式 記為O’分別對YcbCr色彩格式的每個channel經過取樣因 子為Μ的降低取樣後，所得到的大小為（L/M)x(L/M)的影像 記為心。在本節裡由於每個色彩channel内插做法皆相同， 為了方便起見我們只說明Ychannel的做法，灰色多項式内 插法實行步驟描述如下： 步驟1.將影像水平掃描線上的資料取L+1點為一分 段’且Θ—分段的每L+i點資料設為下一分段的第1點資 料，每一分段的資料記為x = ⑷，+ 。 步驟2.將尤經一次累加生成如下： 10 200929894 - x(1)⑷=Σλ⑺ （7) _ for l<k<L + i . ，=1 步驟3.⑯作）代人—維多項式内插法求得内插 值义⑴(仓+ 1/竭。 步驟4.經過-次反累加生成求得估計值你+ 1/竭 x(k + l/M) = [fW (k + i/M)_ i(i) {k)] χ M (8) 0 5.經"2濾波器修正内插估計求得最後結果如下 x(k + 1/M) = 〇x(k) + (1- ay^k + ^ 9 ^ 其中 ’ 0<a<l . 步驟6.依照水平方向内插方法求得垂直方向的内插估 計值。 值件注意的S，在垂直方向的内插方面，如果沒有原始 資料則將水平方向所得的内插值當成原始值計算。 ° 由前述可知，視訊晝面原本有則個晝面’將其降低 取樣為射1個奇數畫面’如此壓縮時只要壓縮射1個奇數書 面，傳輸時也只需傳輸N+1個奇數晝面，與未處理前必項壓 縮咖個晝面以及傳輸則個畫面相較，明顯能夠減少壓 ㈣需要的記憶體容量，且提升傳輸的速度及頻寬利用率； 且在該接收位置B接收到N+1個奇數晝面後，又重 個捨棄的晝面，將視訊畫面恢復為2N+1個 —叫’田於N個 11 200929894 重建晝面都是由相鄰奇數晝面A” A”心 多項式_法话纤得出，故視訊晝面尹物體7、以灰色 物歧的移動更為平順 的移動向旦/:姆以在般必須找出較佳且具有意義 傳輪也相當適合。 更為快速，即使應用在即時視訊重建 Reconstruction step: Estimate the video plane discarded in the sampling step by using the gray polynomial interpolation method. In this embodiment, as shown in Fig. 4, the (four) odd number = surface is used to make the limb color polynomial _ method, and the two phases are estimated. The facets A2, A4, Ae...A2N between the odd-numbered faces restore the video face to 2N+1 facets. The technical description of the gray polynomial interpolation is as follows: (), '隹夕Polynomial inteipolation (PI) Given the original breeding χ(々), the one-dimensional polynomial interpolation method is as follows: Step 1. Let x (the illusion has an L-order polynomial, that is, x(k) ) = cLkL + clJcla + ... + Clk + Cq (1) Step 2. Substituting + 1 into (丨) gives x = Vc (2) where X, C and F represent x (free), 4 and = 0, respectively. y For 0 printed z, + 1. Step 3. Solve the interpolated value + i/w) ^k + VM) = cL(k + l/M)L +--- + Cl(k + l/M) + c0 (3) 8 200929894 where the coefficient heart is obtained from equation (2). Step 4. Obtain the final interpolated value as follows x(k + 1/M) = x(k + l/M) χ Μ ( 4 ) where # denotes an amplification factor and is generally an integer. The randomness between data in one-dimensional polynomial interpolation is one of the main factors affecting its performance. Therefore, if the randomness of the data can be reduced, its performance should be effectively improved. (2) Gray Polynomial Interpolation The grey system is a theory proposed by Professor Deng Julong in 1982, which is based on the gray model GM(1,1). Gray-generated methods are used in the GM(1, 1) model to reduce the randomness between the original data series. The common ash generation method is described as follows: first-order ◎ accumulated generating operation (1-AG0) and first-order inverse accumulated generating operation (1-IAGO). Give the original data 2 0, K h rule}, generate new data according to one accumulation (χ) (9) as follows: x{l\k) = Y^x{n) (5) /7=1 for KB[ 'where χ(ι)(1)(1). From equation (5), you can retrieve the original data from X(1)(/c) easily by 2009 9894. χ(^) = χ^(^)-χ(1)(^-1) (6) for [ . This operation is called an inverse accumulation generation. Here, we propose a grey polynomial interpolation (GPI) to improve the performance of polynomial interpolation. Since the accumulation generation can effectively reduce the randomness between data, in the proposed method, we use the technique of cumulative generation to pre-process the original data, thereby reducing the randomness of the data. The interpolated values are then obtained using one-dimensional polynomial interpolation. The interpolation value is obtained by a post-accumulation post-processing. Finally, an interpolation filter is used to correct the interpolated values. The MXM is the horizontal and vertical magnifications of Μ times, the known size is lxl and the color format is the original image composed of RGB two primary colors, and the color is converted into YCbCr format and recorded as O' for each of the YcbCr color formats. After the channel is sampled with a sampling factor of Μ, the resulting image of (L/M)x(L/M) is recorded as a heart. In this section, since each color channel interpolation method is the same, for the sake of convenience, we only explain the Ychannel approach. The gray polynomial interpolation method is described as follows: Step 1. Take the L+1 point of the image on the image horizontal scan line. The data for each L+i point of a segment 'and Θ-segment is set as the first point data of the next segment, and the data of each segment is recorded as x = (4), + . Step 2. The special summation is generated as follows: 10 200929894 - x(1)(4)=Σλ(7) (7) _ for l<k<L + i . ,=1 Step 3.16) Generation-dimensional polynomial interpolation The method obtains the interpolation value (1) (cage + 1/exhaustion. Step 4. After the -reverse accumulation generation to obtain the estimated value you + 1 / exhaust x (k + l / M) = [fW (k + i / M) _ i(i) {k)] χ M (8) 0 5. After the "2 filter correction interpolation estimate, the final result is as follows x(k + 1/M) = 〇x(k) + (1- Ay^k + ^ 9 ^ where ' 0<a<l . Step 6. Obtain the interpolation estimate in the vertical direction according to the horizontal interpolation method. Note the value of S, in the vertical interpolation, if not The original data is calculated as the original value of the interpolated value in the horizontal direction. ° As can be seen from the above, the video camera originally has a facet 'sampling it to shoot an odd picture' so that as long as the compression is compressed by an odd number In writing, only N+1 odd-numbered faces are transmitted during transmission. Compared with the uncompressed pre-compressed coffee face and the transmitted picture, it is obviously able to reduce the memory capacity required by the pressure (4) and increase the transmission speed. and Wide utilization; and after receiving N+1 odd-numbered faces at the receiving position B, and again discarding the facets, the video images are restored to 2N+1--called 'Tian N' 11 200929894 reconstruction face They are all derived from the adjacent odd-numbered face A" A" heart polynomial _ 话 纤 fiber, so the video face yin object 7, the movement of the gray object is more smooth movement to the Dan /: Better and meaningful transfer is also quite suitable. Faster, even when applied to instant video
〇 12 200929894 【圖式簡單說明】 第1圖本發明實施例取樣步驟的示意圖。 第2圖本發明實施例壓縮步驟的示意圖。 第3圖本發明實施例解碼步驟的示意圖。 第4圖本發明實施例重建步驟的示意圖。〇 12 200929894 [Simple description of the drawings] Fig. 1 is a schematic view showing the sampling procedure of the embodiment of the present invention. Fig. 2 is a schematic view showing a compression step of an embodiment of the present invention. Figure 3 is a schematic diagram of the decoding steps of an embodiment of the present invention. Figure 4 is a schematic illustration of the reconstruction steps of an embodiment of the invention.
【主要元件符號說明】 《本發明》 奇數畫面 Al、八3、Αδ···Α2Ν-1、A2N+1 偶數晝面Α2、Α·4、Αβ…Α2Ν 晝面Α2、人4、Αδ…Α2Ν 接收位置Β 13[Explanation of main component symbols] "Invention" odd-numbered pictures Al, 八3, Αδ···Α2Ν-1, A2N+1 even-numbered Α2, Α·4, Αβ...Α2Ν Α Α2, human 4, Αδ...Α2Ν Receiving locationΒ 13