1270298 p502-005 九、發明說明·· 【發明所屬之技術領域】 本案為一種減少區塊效應及波紋雜訊的方法，尤指一種應用 於影像壓縮之減少區塊效應及波紋雜訊的方法。 【先前技術】 鲁 MPEG是動癌、圖像專豕組(Moving Pictures Experts Group)的英 文^寫，由 ISO(Intemational Standards Organization)與國際電子委 員會(IEC)聯合成立，致力於動態圖像及語音編碼標準化的工作， ， 1998年推出了 MPEG4，其採用國際化的標準，兼容性好，同時提 供更好的壓縮比。 MPEG4將终多的多媒體應用集成一個完整的框架内，用於實 現音視頻(Audio-Video)數據的有效編碼及更為靈活的存取，其書 • 質近乎DVD，但壓縮比遠高與DVD。簡單的說，MPEG4具有高 壓縮比、節省存儲及畫質佳等優點。 在MPEG-4規格書AnnexF中提出解區塊效應(debl〇cking) 和解波紋雜訊(deringing)的做法。其中，debl〇cking需要對影像區 塊k界（block boundary)兩旁各4個像素(pixeis)來判斷是否有 棱邊(edge)，並且使用較耗費計算量的方式來處理。而 的做法需先求出整個影像區塊(bl〇ck)的最大及最小值，然後設定 臨界值(threshold)來判斷block裡每個pixd是否需要做 1270298 p502-〇〇5 deringing，如果需要，則用一個權重遮罩(wdghted mask)來整平 (smooth)此像素(pixel)。 習用技術中’ H.264的deblocking做法，則是根據相鄰兩個 blocks之間的關係’來判斷其邊界力量(b〇undary 。 boundary strength越強，則有blocking的可能性越高。可能性越 高，則採用越強的濾波(filtering)方法，而其強度總共分為〇〜5 類，因此其方法比MPEG-4的做法更為複雜，並且此種做法只能 # 與H.264的壓縮機制搭配。H.264則無提供deringing的做法。 其次，如美國專利號6,259,823、5,883,983、5,877,813等相關 -習用技術’皆為同一個作者所申請，其方法皆類似：先對影像求 '得全域(§lobal)以及區域(1〇cal)稜邊圖(edge map)，然後得到二進位 ，稜邊圖(binary edge map)。根據此binaiy edge map，判斷每一像素 (pixel)在稜邊(edge)上或是在平坦區域上，然後使用不同的 # 版來處理。但是，習用的作法複雜，所使用的硬體成 本高，且速度較慢。 【發明内容】 本案的目的絲據習用技術的缺點，提出—新穎進步實用的 改善方^’使Deblocking與Deringing絲更關單，降低所使用 的硬體成本，且可提昇運算的速度。 為達上述目的，本案提出—種顧於影像壓縮之減少區塊效 應及波紋雜訊的方法，係針對—影像之每—影像區塊，除了最後 1270298 P502-005 一列之影像區塊與最後一行之影像區塊，由左而右，由上而下， 做下列之處理： 對於每區塊列(blockrow)，由左而右，對其中之每一影像區塊做處 理，其中： 對於一區塊而言，因應每一像素列(pixel row)之邊界(b〇uncjary)為 非稜邊(edge)，修改跨於該邊界之二個像素(pixei)的值； 因應低活動(lowactivity)之像素列，對其中每一像素做較強烈的整 平(smooth)動作；而 因應非低活動之像素列，對其中具有相近值的相鄰像素做整平的 動作。 如所述之應用於影像壓縮之減少區塊效應及波紋雜訊的方 法，其中更對於母區塊行(block column)，由上而下，對其中之影 像區塊做處理，其中： 對於區塊而θ目應每一像素行(pixel c〇lu·)之邊界作_切) 為非稜邊(edge)，修改跨於該邊界之二個像素(pixd)的值； 因應低活動(l0W activity)之像素行，對其中每一像素做較強烈的整 平(smooth)動作；而 口應非低雜之像素仃，對其中具有相近值的相鄰像素做整平的 動作。 如所述之應用於影像壓縮之減少區塊效應及波紋雜訊的方 法’其中該影像係為-解壓縮後之影像，可應用於鹏仏 1270298 P502-005 MPEG-2、MPEG-4、以及H.264等以影像區塊為基礎之壓縮機制 之迴圈中(in-loop)或後處理(post-processing)階段。 如所述之應用於影像壓縮之減少區塊效應及波紋雜訊的方 法，其中該影像區塊的大小係由所採用之壓縮機制之内建參數決 定。 如所述之應用於影像壓縮之減少區塊效應及波紋雜訊的方 法，其中該低活動係指一像素列(或一像素行)中像素間的值的差距 較小的情況。 【實施方式】 圖l(AXB)為deblocking和deringing應用於網際網路及一般電 腦的示意圖。圖1(A)中，指出使用影像區塊方式壓縮影像可供資 料傳輸，經由網路到使用者的電腦，接收資料後接著進行解壓縮， 而後可進行本案所提之deblocking及deringing的動作，然後顯示 在顯示器或螢幕上面。圖1(B)中，壓縮的影音資料係存放在電腦 的儲存裝置（例如硬碟）内。在播放時，直接進行解壓縮，而後 同樣可進行本案所提之deblocking及deringing的動作，然後顯示 在顯示器或螢幕上面。 何謂 deblocking ? 區塊效應(blocking)係由於壓縮技術所致，以區塊方式(B1〇ck based)壓縮的技術，例如DCT等，都會在壓縮時產生某些程度的 1270298 P502-005 區塊效應。 區塊效應通常會有明顯的邊界(Boundary)，尤其在一平坦的區 域中，格外明顯。 因此，當影音資料解壓縮後即必須進行debl〇ckin^々處理， 將會產生區塊效應的區塊邊界整平同時維持稜邊(以與)的完好無 缺0 圖2(A)(B)(C)舉出區塊效應的三種形式。圖2(a)為階梯形 (staircase)的區塊2丨，在影像的邊緣地方產生階梯狀的圖樣。圖2毋)1270298 p502-005 IX. INSTRUCTIONS · · Technical Fields of the Invention The present invention is a method for reducing blockiness and ripple noise, and more particularly to a method for reducing blockiness and ripple noise applied to image compression. [Prior Art] Lu MPEG is an English-language writing of the Moving Pictures Experts Group. It is jointly established by ISO (Intemational Standards Organization) and the International Electrotechnical Commission (IEC). It is dedicated to dynamic images and speech. In the work of coding standardization, MPEG4 was introduced in 1998, which adopts international standards, has good compatibility, and provides better compression ratio. MPEG4 integrates the most multimedia applications into a complete framework for efficient encoding and more flexible access of audio-video data. The book quality is close to DVD, but the compression ratio is much higher than DVD. . Simply put, MPEG4 has the advantages of high compression ratio, saving storage and good picture quality. The practice of deblocking (declining) and de-coring noise is proposed in the MPEG-4 specification Annex F. Among them, debl〇cking needs to determine whether there are edges (pixeis) on both sides of the block boundary of the image block, and it is processed in a more computationally expensive manner. The first method is to find the maximum and minimum values of the entire image block (bl〇ck), and then set the threshold (threshold) to determine whether each pixd in the block needs to do 1270298 p502-〇〇5 deringing, if necessary, Then use a weight mask (wdghted mask) to smooth the pixel. In the conventional technology, the 'deblocking method of H.264 is to judge the boundary strength according to the relationship between two adjacent blocks' (b〇undary. The stronger the boundary strength, the higher the possibility of blocking. Possibility The higher the filter, the stronger the filtering method, and the strength is divided into 〇~5, so the method is more complicated than MPEG-4, and this method can only be ## with H.264 The compression mechanism is matched. H.264 does not provide the practice of deringing. Secondly, such as US Patent Nos. 6,259,823, 5,883,983, 5,877,813 and other related-practical technologies are all applied by the same author, and the methods are similar: Global (§lobal) and region (1〇cal) edge map, and then get the binary edge map. According to the binaiy edge map, judge each pixel (pixel) on the edge (edge) or on a flat area, and then use a different # version to deal with. However, the practice is complicated, the hardware used is costly, and the speed is slow. [Summary of the invention] Technical shortcomings, The new-progressive and practical improvement method makes Deblocking and Deringing more relevant, reduces the hardware cost used, and can improve the speed of calculation. To achieve the above purpose, this case proposes a reduction zone for image compression. The method of blockiness and ripple noise is for each image block of the image, except for the image block of the last row of 1270298 P502-005 and the image block of the last row, from left to right, from top to bottom, The following processing: For each block row, from left to right, each image block is processed, where: For a block, corresponding to the boundary of each pixel row (pixel row) B〇uncjary) is a non-edge, modifying the value of two pixels (pixei) across the boundary; for each pixel of the low activity, a stronger leveling (smooth) Action; and in response to non-low activity pixel columns, the adjacent pixels with similar values are leveled. As described above, the method of reducing blockiness and ripple noise is applied to image compression, and more Parent block line (blo Ck column), from top to bottom, the image block is processed, where: for the block and θ should be the edge of each pixel row (pixel c〇lu·) _cut) is non-edge (edge ), modifying the value of the two pixels (pixd) across the boundary; in response to the pixel row of the low activity (l0W activity), a stronger smoothing action is performed on each of the pixels; and the mouth should be non-low noise The pixel 仃 is an action of leveling adjacent pixels having similar values. As described in the method of reducing blockiness and ripple noise applied to image compression, wherein the image is a decompressed image, which can be applied to Pengyi 1270298 P502-005 MPEG-2, MPEG-4, and An in-loop or post-processing phase of an image block-based compression mechanism such as H.264. As described, the method for reducing blockiness and ripple noise applied to image compression, wherein the size of the image block is determined by the built-in parameters of the compression mechanism employed. As described, the method for reducing blockiness and ripple noise applied to image compression, wherein the low activity refers to a case where the difference in values between pixels in a pixel column (or a pixel row) is small. [Embodiment] FIG. 1 (AXB) is a schematic diagram of deblocking and deringing applied to the Internet and a general computer. In Fig. 1(A), it is pointed out that the image block is used to compress the image for data transmission, and the data is transmitted to the user's computer via the network, and then the data is received and then decompressed, and then the deblocking and deringing actions mentioned in the present case can be performed. It is then displayed on the display or screen. In Fig. 1(B), the compressed video data is stored in a storage device (such as a hard disk) of the computer. During playback, the decompression is directly performed, and then the deblocking and deringing actions mentioned in the present case can be performed, and then displayed on the display or the screen. What is deblocking? Blocking is due to compression techniques. Techniques that are compressed in a block mode, such as DCT, will produce some degree of 1270298 P502-005 block effect when compressed. . Block effects usually have significant boundaries (Boundary), especially in a flat area, which is particularly noticeable. Therefore, when the audio-visual data is decompressed, it must be debl〇ckin^々, which will result in block-level block boundary leveling while maintaining the edge (and) intact. Figure 2(A)(B) (C) Three forms of blockiness are cited. Fig. 2(a) shows the block 2丨 of the staircase, which produces a stepped pattern at the edge of the image. Figure 2毋)
為格狀(grid)區塊22，在影像平坦的區域產生格狀的圖樣。圖2(Q 為角落凸起23(Comer outlier)的圖樣。這三種形式的區塊效應如果 /又去除或去除的不好，影像則會變的模糊不清。圖3為影像具有 區塊效應的例子。 何謂 de-ringing? 影像之波紋雜訊(ringing)係由尖銳的切斷(cut _〇ff)濾波器所產 生’例如濾波器只選擇了低頻成份，在量化(quantizati〇n)的過程中 、。成了解析度的p牛低。波紋雜訊(ringing)係為棱邊(吨e)附近的失 真(Aliasing)現象’尤其是稜邊(edge)圍繞著平坦區域時最容易產 生。清除這種波紋雜訊(ringing)的方法即為。圖4為影像 具有波紋雜訊的例子。 de七lockmg及de_ringing可以有效及確實地減少影像看起來不 自…、的現1叫亦可維持影像的完整性。其困難在於，如何從 1270298 ^ . P502-005 、、” ’ σ為相鄰二像素差的絕對值，用來判斷二像素之間的 活動力(actlvity)。目河區塊61的像素以表示，右 像素以坤表示，其步驟如下： 兒62的 1:係修改跨於邊界二像素（例如C18及R11二點)的值。如果 跨於邊界二像素的值相減後為泊，小於零，且在負祕的範圍 内=將d3加上△；!數列中最小的值，並暫存於t;如果 大於令’且在正threl的範圍内，則Μ減去△數列甲最小的值 (η(Λι)) ’暫存於t。然後將t右移齡2個位元加2)，並回 存於t。取後，再將Ci8減去t並回存於α8 ; Ril加上t並回存於 Ril 〇 ^ 2:檢查第i列的活動(Activity): 如果數列Ai中最大值小於thre3，則跳到步驟4 ;否則，設定 計數裔cnt=0,針對每個(jj，於〇*j大於thre4時，計數器cnt則累 加1。如果到最後，cnt小於thre5，則用5-tapU>F來整平像素列 中的每個像素(Cil到Ci8)，然後跳至步驟4。 3:整平Cil到Ci8與其左或/及右邊的像素： 0又dl為目鈾運异像素與其左邊像素的差的絕對值，似為目前 運算像素與其右邊像素的差的絕對值； 如果dl小於thre6且d2也小於thre6，則整平目前的像素(aj) 與其左右相鄰二點的像素； 否則，如果dl小於thre6，則整平目前的像素(Cij)與其左相鄰In the grid block 22, a grid pattern is created in a region where the image is flat. Figure 2 (Q is the pattern of the corner protrusion 23 (Comer outlier). If the three forms of block effect are not removed or removed, the image will become blurred. Figure 3 shows the image has a block effect. The example of what is de-ringing? The imagery of the ringing is produced by a sharp cut-off filter. For example, the filter only selects the low-frequency component, which is quantized (quantizati〇n). In the process, the resolution is low, and the ringing is the distortion phenomenon near the edge (ton e), especially when the edge surrounds the flat area. The method of clearing such ripple noise is as shown in Fig. 4. The image has corrugated noise. The de seven lockmg and de_ringing can effectively and surely reduce the appearance of the image. To maintain the integrity of the image, the difficulty lies in how to determine the absolute value of the adjacent two pixel difference from 1270298 ^ . P502-005 , , ' σ is used to determine the activity between two pixels. The pixel of block 61 is represented by the right pixel, and the steps are as follows. Bottom: 1:2 of the child 62 is modified to cross the value of two pixels across the boundary (for example, two points of C18 and R11). If the value of two pixels across the boundary is subtracted, it is a mooring, less than zero, and within the range of negative secret = Add d3 to the smallest value in the △;! sequence and temporarily store it in t; if it is greater than the order 'and in the range of positive threl, then subtract the smallest value of the △ series A (η(Λι)) 'temporary Then t. Then shift t to the next 2 bits and add 2) and return to t. After taking it, subtract Ci from t8 and return it to α8; Ril plus t and return to Ril 〇^ 2 : Check the activity of the i-th column: If the maximum value in the series Ai is less than thre3, skip to step 4; otherwise, set the count cnt=0, for each (jj, when 〇*j is greater than thre4, the counter Cnt accumulates 1. If cnt is less than thre5 at the end, use 5-tapU>F to level each pixel in the pixel column (Cil to Ci8), then skip to step 4. 3: Leveling Cil to Ci8 and its Left or / and right pixels: 0 and dl is the absolute value of the difference between the uranium pixel and its left pixel, which is the absolute value of the difference between the current operation pixel and its right pixel; if dl is less than thre6 and d2 In thre6, leveling the current pixel (AJ) around its two points of adjacent pixels; otherwise, if dl is less than thre6, leveling the current pixel (Cij is) left adjacent thereto
11 (S 1270298 p502-005 點的像素； 否則，如果d2小於thre6，則整平目前的像素(cij)與其右相鄰 點的像素。 4:做下一列。 以上是針對列的演算法，行的演算法則同理。 本案的deblocking和deringing的整合性做法比MpEG4 鲁 簡單許多，在速度上比MPEG-4快許多（約3〜4倍），並且效果 不輪給MPEG-4。下表所示，共實驗了 117張影像，其中之數字 ^其平均值，SNR為影像品質的一種量化衡量法，值越高品質越 . it而TimeSpent為所花時間，越少越快。avg為average之意。 本案 MPEG-411 (S 1270298 p502-005 points of pixels; otherwise, if d2 is smaller than thre6, then level the current pixel (cij) with the pixel of its right neighbor. 4: Do the next column. The above is the algorithm for the column, the line The algorithm is the same. The integration of deblocking and deringing in this case is much simpler than MpEG4, which is much faster (about 3 to 4 times) than MPEG-4, and the effect is not round to MPEG-4. A total of 117 images were tested, of which the average value, SNR is a quantitative measure of image quality. The higher the value, the higher the quality. It and TimeSpent are the time spent, the less the faster. The avg is average This case is MPEG-4
avgSNR(平均訊號雜訊比） 10.7301 dBavgSNR (average signal noise ratio) 10.7301 dB
10.7200 dB rgTimeSpent(平均時間） 4.2697 秒 11.4610 秒 本案的做法不·定與Η·264的壓誠健配，並且比 Η·264的deblocking做法簡單許多。 一 卜本案所採用的方法，不須建立global/localedgemap， :卻了這兩部份儲存空間。而在判斷—個pixel處在_或是 ^區域上的做法，以及所_的棘_牡，本細故法都比 季父間易。 ^270298 p502-005 本案所揭露之技術，得岭習本麟人士據以實施，而其前10.7200 dB rgTimeSpent (average time) 4.2697 seconds 11.4610 seconds The practice of this case is not the same as that of Η·264, and it is much simpler than the deblocking method of Η·264. The method used in this case does not need to establish a global/localedgemap: but these two parts of the storage space. In the judgment of a pixel in the _ or ^ area of the practice, as well as the _ _ _ _ _, this fine method is easier than the season father. ^270298 p502-005 The technology disclosed in this case, the person of Deling Xiben Lin was implemented, and before
:未^料亦專雜，爰錄提料狀 _例尚不心X喊本賴欲健 U 專利範圍如附。 引乾圍，因此，提出申請: Unexpected material is also special, 爰 提 提 提 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Leading the application, therefore, apply
13 l27〇298 P502-005 【圖式簡單說明】 圖1(A)(B)為本案deblocking和deringing應用於網際網路及一般電 腦的示意圖。 圖2(A)(B)(C)為本案舉出區塊效應的三種形式。 圖3為本案影像具有區塊效應的例子。 圖4為本案影像具有波紋雜訊的例子。 圖5為本案de-blocking及de-ringing的整體做法。 圖6為本案詳細演算法。 【主要元件符號說明】 階梯形區塊.........21 格狀區塊..........22 角落凸起..........23 解壓縮後的影像.......51 影像區塊..........511 目前的區塊.........52 右方區塊..........53 下方區塊..........54 最後一行及最後一列的區塊·· 512 目前區塊..........61 右邊區塊..........62 目前區塊的像素.......Cij 1270298 p502-005 右邊區塊的像素.......Rij13 l27〇298 P502-005 [Simple description of the diagram] Figure 1 (A) (B) is a schematic diagram of deblocking and deringing applied to the Internet and general computers. Figure 2 (A) (B) (C) shows three forms of block effect in this case. Figure 3 shows an example of the block effect of the image in this case. Figure 4 shows an example of the image with ripple noise. Figure 5 shows the overall practice of de-blocking and de-ringing. Figure 6 shows the detailed algorithm of this case. [Description of main component symbols] Stepped block.........21 Lattice block..........22 Corner bump..........23 Solution Compressed image.......51 Image block..........511 Current block.........52 Right block... ....53 Lower Blocks..........54 Last and Last Columns·· 512 Current Blocks..........61 Right Blocks... .......62 The pixel of the current block....Cij 1270298 p502-005 The pixel of the right block....Rij
dd