• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Screen Content Coding with Primary and Secondary Reference Buffers for String Matching and Copying

    2015-10-11 07:02:53TaoLinKailunZhouandLipingZhao
    ZTE Communications 2015年4期

    Tao Lin,Kailun Zhou,and Liping Zhao

    (Tongji University,Shanghai 200092,China)

    Screen Content Coding with Primary and Secondary Reference Buffers for String Matching and Copying

    Tao Lin,Kailun Zhou,and Liping Zhao

    (Tongji University,Shanghai 200092,China)

    A screen content coding(SCC)algorithm that uses a primary reference buffer(PRB)and a secondary reference buffer(SRB)for string matching and string copying is proposed. PRB is typically the traditional reconstructed picture buffer which provides

    tring pixels for the current pixels being coded.SRB stores a few of recently and frequently ref?erenced pixels for repetitive reference by the current pixels being coded.In the encoder,searching of optimal reference string is performed in both PRB and SRB,and either a PRB or SRB string is selected as an optimal reference string on a string?by?string basis.Compared with HM?16.4+SCM?40 refer?ence software,the proposed SCC algorithm can improve cod?ing performance measured by bit?distortion rate reduction of average 4.19%in all?intra configuration for text and graphics with motion category of test sequences defined by JCT?VC common test condition.

    HEVC;Image Coding;Screen Content Coding;String Match?ing;Video Coding

    1 Introduction

    W ith continuous and rapid advancements in communications,networking,computers,semi?conductors,and displays technologies,screen content coding(SCC)becomes a key technolo?gy for some emerging popular applications such as cloud com?puting,cloud?mobile computing,Wi?Fi display,smartphone and tablet second display,ultra?thin clients,remote desktops,and screen sharing[1]-[3].The challenging requirement of SCC is to achieve both ultra?high visually lossless quality and ultra?high compression ratio up to 300:1-3000:1.In recent years,SCC has attracted increasing attention of researchers from both academia and industry[4]-[30].ISO/IEC MPEG and ITU?T VCEG have set up the Joint Collaborative Team(JCT)for a joint project of developing a SCC version of High Efficien?cy Video Coding(HEVC,also known as H.265)standard and a joint call?for?proposal has been issued in January 2014[14].

    Characteristics of computer screen pictures are quite differ?ent from those of traditional pictures,for which traditional block?matching and transform?based hybrid coding technique can compress efficiently.In general,computer screen pictures are essentially a mix of two?type contents:discontinuous?tone content such as text,chart,graphics,and icon,and continuous?tone content such as photograph,movie/TV clips,natural pic?ture video sequences,and sophisticated light?shaded and tex?ture?mapped virtual?reality scenes.Continuous?tone content features relatively smooth edges,complicated textures,and thick lines with virtually unlimited colors.In contrast,discon?tinuous?tone content features very sharp edges,uncomplicated shapes,and thin lines with few colors,even one?pixel?wide sin?gle?color lines.General screen content pictures are often seen in web browsing,video conferencing with document sharing,of?fice document editing,engineering drawing,hardware design engineering,software programming,gaming,map navigating,address direction searching,and other computer use cases.

    Actually,typical computer screens for daily use are often rich in small and sharp bitmap structures such as text,menu,icon,button,slide?bar,and grid.There are usually many simi?lar or identical patterns in a screen picture.These similar or identical patterns may have very different sizes from a few pix?els to a few thousands of pixels and very different shapes.Ex?isting techniques such as intra?prediction and intra block copy(IBC)[15]-[17]is not always efficient to code similar or identi?cal pattern within a picture,because they use only 1?D pattern or 2?D pattern of a few fixed sizes and a few fixed rectangle or square shapes.

    In order to address the issue and improve the coding effi?ciency of repeated patterns with different sizes and shapes,this paper proposes a coding technique based on string?match?ing(also called intra string copy or ISC).It provides a very ge?neric and flexible solution to string matching of variable sizes and different shapes.

    Two reference buffers are used in the proposed technique. One is primary reference buffer(PRB)that is typically the tra?ditional reconstructed picture buffer to provide reference string pixels for the current pixels being coded.The other is second?ary reference buffer(SRB),a lookup table(LUT)storing a few of recently and frequently referenced pixels for repetitive refer?ence by the current pixels being coded.For any starting pixel in an encoding process,searching of optimal reference string is performed in both PRB and SRB.Either a PRB string or a SRBstring is selected as an optimal reference string on a string?by?string basis.A PRB string has two string?matching parameters. One is an offset that specifies the relative position(2D coordi?nates),i.e.2D displacement from the reference string to the current string.The other is a length that specifies the number of pixels in the reference string.It is obvious that the reference string and the current string have the same length.A SRB string also has two string?matching parameters:an index that specifies the address of the reference pixel in the SRB and a length that specifies the duplication count of the reference pix?el to reconstruct the current string.The string?matching param?eters are then entropy?coded and put into the final bit?stream. At the decoder side,the string?matching parameters are actual?ly string?copying parameters.In the decoding process,the string?matching(i.e.string?copying)parameters are parsed,de?coded,and used to obtain the reference string pixels from ei?ther PRB or SRB.The values of the reference string pixels are then assigned to the current pixels being decoded to recon?struct the current string.

    In Section 2,a general architecture of string?matching en?hanced coding system using both PRB and SRB is proposed and the details of horizontally and vertically scanned 2D?shape?preserved matching modes are described.Section 3 presents a flexible hash?table framework specially designed for speeding up reference string searching in PRB.Section 4 describes a method to seamlessly mix PRB search with SRB search inside a coding unit(CU),and to select one?by?one either an optimal PRB string or an optimal SRB string based on average rate?dis?tortion(RD)cost evaluation.In section 5,the proposed tech?nique is compared with HM?16.4+SCM?40 reference software[31]that is developed based on HEVC Software Model HM?16.4 by JCT for SCC testing.The experimental results show that the proposed technique can achieve significant bit?distor?tion rate(BD?rate)[32]-[33]reduction in lossy coding and bit?rate saving in lossless coding using the test suite defined by JCT Common Test Condition[30].We conclude the paper and introduce the future work on string?matching coding technique in section 6.In this paper,the proposed SCC algorithm is de?scribed for pictures of YUV4:4:4 and RGB formats,but it can also be applied to pictures of YUV4:2:2 and YUV4:2:0 formats with some modifications.This paper is partially based on JCT?video coding(VC)documents[18]-[27].

    2 String-Matching Enhanced Coding System

    A general architecture and major components of string?matching enhanced coding system are shown in Fig.1.

    In the encoder of the string?matching enhanced coding sys?tem,the string?matching encoding subsystem performs color clustering,PRB string searching,and SRB string searching. The rest of the encoding system performs other traditional en?coding operations such as intra?prediction,inter?prediction,transform,quantization,entropy encoding,IBC,and palette coding.The input CU O is fed to both string?matching encod?ing subsystem and the rest of the encoding system.The string?matching encoding subsystem codes O to generate coded bit?stream b1and reconstructed CU P1.The rest of the encoding system also codes O to generate coded bit-stream b2and recon?structed CU P2.O,P1,b1,P2,and b2are sent to RD?cost?based selector that calculates the RD cost of two results and selects the one with the best RD performance as the final coding result for the CU.The corresponding coded bit-stream b1or b2is se?lected and put into the output bit-stream.Also,the correspond?ing reconstructed CU P1or P2is stored in reconstructed picture buffer,which is shared by both string?matching encoding sub?system as PRB for string?matching and the rest of the encoding system for inter prediction and IBC.The input CU O is also fed to a color cluster unit to obtain a few of most frequently oc?curred pixel colors.Some of the colors are put into the SRB LUT for SRB string searching.

    In the decoder,the input bit-stream is first parsed by CU cod?ing type parser to get the CU_coding_type_flag that indicates whehter the current CU being decoded is coded by the pro?posed string?matching technique or by traditional coding tech?niques.If the CU is coded by the string?matching technique,the CU layer bit-stream is sent to the string?matching decoding subsystem that decodes and reconstructs the CU P1,which is stored in reconstructed picture buffer and is also the final re?constructed CU output ?.If the CU is coded by traditional cod?ing techniques,the CU layer bit-stream is sent to the rest of the decoding system that performs traditional decoding tasks such as intra?prediction,inter?prediction,inverse?transform,de?quantization,IBC,and palette decoding.The CU P2is eventual?ly reconstructed,which is stored in reconstructed picture buf?fer and is also the final reconstructed CU output ?.

    The string?matching coding subsystem has two CU level matching modes:horizontally scanned 2D?shape?preserved matching and vertically scanned 2D?shape?preserved matching(Fig.2).A CU is pre?coded by both the modes.The mode mini?mizing RD cost is selected as the final string?matching mode for the CU.The CU size in Fig.2 is 16x16 pixels in the packed pixel format(e.g.a Y sample is followed by a U sample and then a V sample,or a G sample is followed by a B sample and then an R sample).

    The vertically scanned 2D?shape?preserved matching mode is used to code CU m in Fig.2.In this mode,PRB is treated as a 2D plane while CU m is considered a vertically scanned 1D pixel string.Starting from the 1st pixel of CU m,the encoder searches the optimal(e.g.longest in lossless case)string in the PRB searching range that matches the pixel string in CU and also keeps exactly the same 2D shape as the pixel string in CU m.The string found in the searching range is called a refer?ence string and the pixel string in CU m is called current string.Fig.2 shows the first two current strings in CU m and their corresponding reference strings:

    ▲Figure 1.String?matching enhanced coding system architecture.

    ▲Figure 2.Two matching modes of string?matching coding.

    1)The 1st reference(and current)string in yellow has 15 pix? els and the 2D?shape?preserved reference string is located across CU 1 and CU h+1;

    2)The 2nd reference(and current)string in cyan has 20 pixels and the 2D?shape?preserved reference string is located across CU 1,CU 2,and CU h+1.

    The horizontally scanned 2D?shape?preserved matching mode is used to code CU m+1 in Fig.2.In this mode,PRB is treated as a 2D plane while CU m+1 is considered to be a hori? zontally scanned 1D pixel string.Starting from the 1st pixel of CU m+1,the encoder searches the optimal string in the PRB searching range that matches the pixel string in CU m+1 and also keeps exactly the same 2D shape as the pixel string in CU m+1.The string found in the searching range is called a refer?ence string and the pixel string in CU m+1 is called the cur?rent string.Fig.2 shows the first three current strings in CU m+ 1 and their corresponding reference strings:

    1)The 1st reference(and current)string in pink has 24 pixels and the 2D?shape?preserved reference string is located across CU 1 and CU 2;

    2)The 2nd reference(and current)string in black has 15 pix?els and the 2D?shape?preserved reference string is located across CU h and CU h+1;

    3)The 3rd reference(and current)string in orange has 18 pix?els and the 2D?shape?preserved reference string is located across CU 1 and CU 2.

    There are usually two types of pixel scanning methods(and paths)available for a CU or a largest coding unit(LCU).One is raster?scan:a CU or LCU is scanned line by line,either hori?zontally or vertically,and all lines have the same scan direc?tion(Fig.2).The other is traverse?scan:a CU or LCU is also scanned line by line,either horizontally or vertically,but odd lines and even lines have opposite scan direction.Once a scan?ning method(and path)is determined,all pixels inside a CU or LCU are lined up following the scanning path.Thus,starting from a current pixel being coded inside a CU Pm,n,a current string curSm,ncan be defined following the scanning path.The pixels of curSm,nare designated as Sm,n(0),Sm,n(1),...,Sm,n(L?1),following the order of the pixels on the scanning path,where L is the length of the string in unit of pixel.Using the designa?tion,curSm,nand its pixels can be expressed as curSm,n={Sm,n(k):0≤k<L}.Given a current string curSm,nand a reference pixel Pi,j,a reference string refSi,j={Si,j(k):0≤k<L}is the string that starts from Pi,j,i.e.Si,j(0)=Pi,j,and has exactly the same 2D shape and scanning order as curSm,n.

    Fig.3 illustrates an example of the current string curSm,n={Sm,n(k):0≤k<L}starting from the current pixel Pm,ninside a 8×8 CU of horizontal traverse?scan and its reference string refSi,j={Si,j(k):0≤k<L}starting from the reference pixel Pi,j.Both curSm,nand refSi,jhave identical 2D shape and length of L=9.

    Obviously,given a current pixel Pm,n,the current string curSm,nof length L can be uniquely specified within a CU with known size and scanning method,and a reference string refSi,jis uniquely specified by a displacement vector(DVh,DVv)=(i?m,j?n).In Fig.3,DVh=i?m is the horizontal displacement be?tween curSm,nand refSi,j,and DVv=j?n is the vertical displace?ment between curSm,nand refSi,j.

    reference string refSi,j={Si,j(k):0≤k<L}for a given current string curSm,n={Sm,n(k):0≤k<L}satisfies the constraint that the difference between Si,j(k)and Sm,n(k)is within a predetermined threshold for all k.One common option used for difference measurement is the absolute difference|Si,j(k)Y?Sm,n(k)Y|,|Si,j(k)U?Sm,n(k)U|,and|Si,j(k)V?Sm,n(k)V|,where the sub?scripts Y,U,V designate the Y color component,U col?or component,and V color component for the corresponding pixels Si,j(k)or Sm,n(k).

    3 String-Matching Oriented Hash-Table Framework

    ▲Figure 3.An example of current string and its reference string.

    String?matching coding performance heavily depends on the PRB searching range.The larger the searching range is,the better the coding performance can achieve.However,the lon? ger the searching time is,the more the computing power re?quires.Hash?table can be used to speed up reference string searching.Therefore,the key to design a practical string?matching encoding subsystem is to have a single and efficient string searching oriented hash?table that should work and speed up the searching in all two matching modes.

    In the string?matching oriented hash?table framework,the basic role of a hash?table is to quickly find the first matching reference pixel in the PRB searching range by table?look?up. First of all,we need to define a pixel?order for all pixels in the PRB searching range.Naturally,we use the order defined in the horizontally scanned string matching mode,that is,all pix?els are ordered one LCU followed by another LCU in LCU cod?ing order.Within an LCU,horizontal scanning is employed to order pixels.All pixels with the same hash?value are chained together,following the pixel?order.The hash?value of any cur?rent pixel being coded is calculated and the encoder only needs to search through the hash chain of the same hash?value,instead of all pixels of the entire PRB searching range,to find a potential matching reference pixel.This potential pixel is the first pixel of a potential reference string.In this way,the searching time can be significantly reduced.

    In Fig.4,the pixels(in the searching range)lined up in pixel?order are P0,0,P1,0,P2,0,P3,0,...,Pi,j,Pi+1,j,Pi+2,j,Pi+3,j,...,Pi,j+1,...,followed by the current pixel being coded,Pm,n.The hash val?ues ha_val0,0,ha_val1,0,ha_val2,0,ha_val3,0,...,ha_vali,j,ha_vali+1,j,ha_vali+2,j,ha_vali+3,j,...,ha_vali,j+1are computed from the corre?sponding pixels.There are three hash chain examples in this figure.

    1)Example 1

    Three pixels P0,0,Pi,j,and Pi+2,jhave the same hash value,i.e. ha_val0,0=ha_vali,j=ha_vali+2,j=x.Coordinates(0,0),(i,j),and(i+2,j)of the three pixels are stored in the hash?table and form a hash chain of hash?value x.The hash?table actually has two parts:chain?head and chain?body.Among the three coordi? nates,the coordinates(i+2,j)having the largest pix?el?order is stored in the chain?head slot of address x.The coordinates(i,j)having the second largest pixel?order is stored in chain?body location of coor?dinates(i+2,j).The coordinates(0,0)having the third largest pixel?order is stored in the chain?body location of coordinates(i,j).Finally,?1 is stored in the chain?body location of coordinates(0,0)to indi?cate the end of the chain.

    2)Example 2

    Two pixels P2,0and Pi+1,jhave the same hash val?ue,i.e.ha_val2,0=ha_vali+1,j=y.Coordinates(2,0)and(i+1,j)form a hash chain of hash?value y in the hash?table.In the hash chain,the coordinates(i+1,j)having the largest pixel?order is stored in the chain?head slot of address y.The coordinates(2,0)having the second largest pixel?order is stored in the chain?body location of coordinates(i+1,j).Finally,?1 isstored in the chain?body location of coordinates(2,0)to indi?cate the end of the chain.

    ▲Figure 4.Hash?table structure and contents.

    3)Example 3

    The third hash chain has only one pixel P1,0whose hash val?ue is ha_val1,0=z.Therefore,the coordinates(1,0)is stored in the chain?head slot of address z and?1 is stored in the chain?body location of coordinates(1,0)to indicate the end of the chain.

    When the current pixel Pm,nis encoded and the current string starts from Pm,n,the hash value ha_valm,nof Pm,nis first computed and then used as the chain?head address to find the 1st coordinates in the hash chain of the same hash value.In the hash chain examples in 0,if ha_valm,n=x,the 1st coordi?nates is(i+2,j);If ha_valm,n=y,the 1st coordinates is(i+1,j);If ha_valm,n=z,the 1st coordinates is(1,0).The content of the 1st coordinates in the chain?body is the 2nd coordinates in the hash chain of the same hash value,and the content of the 2nd coordinates in the chain?body is the 3rd coordinates in the hash chain of the same hash value,and so on.Therefore,the string?matching encoder can use the hash value of the current pixel being coded to find all pixels and their locations that have the same hash value in the searching range.Moreover,a hash chain starts from a hash?head slot,the 1st coordinates have the largest pixel?order,the 2nd coordinates have the sec?ond largest pixel?order,and so on.

    For a pixel P=(Y,U,V)or(G,B,R),where Y,U,V(or G,B,R)are three 8?bit color components of P,a 12?bit hash val? ue ha_val is computed.For lossy coding,ha_val is computed by concatenating 4?bit most significant bit(MSB)of Y,U,V(or G,B,R)to form a 12?bit value.For lossless coding,ha_val is computed by cyclic redundancy check(CRC)?12 algorithm[34]to get a 12?bit value.Obviously,all the pixels are always located in the same hash chain,no matter they have the identi?cal 4?bit MSB component value in a lossy coding case or the identical 8?bit component value in a lossless coding case. Therefore,a hash chain with the same hash value provides a small and efficient candidate set of reference pixels for the starting pixel of a potential reference string.

    4 PRB and SRB Based String-Matching

    In the proposed string?matching technique,a current string being coded gets reference pixels from either PRB or SRB. When a current string gets the reference pixels from PRB,the reference pixels form a reference string that has exactly the same 2D shape and length(number of pixels)as the current string.The reference string can be in any valid location inside PRB.When a current string gets the reference pixels from SRB,it actually gets only one single reference pixel color from SRB LUT and all pixels of the entire string has the same color.

    ▲Figure 5.A CU coded by string?matching technique.

    A CU coded by the string?matching technique can have both PRB and SRB strings(Fig.5).In the figure,the 1st string in purple dots is a 4?pixel PRB string.Its reference string is just above it and has the same horizontal line shape and length of 4pixels.The 2nd string in red dots is a 2?pixel SRB string.Its reference pixels are the 1st SRB LUT pixel color duplicated twice.The 3rd string in purple dots is a 43?pixel PRB string. Its reference string is located across the boundary between cur?rent coding tree unit(CTU)and above CTU.The PRB string and its reference string have exactly the same horizontal tra?verse?scan shape and length of 43 pixels.The 4th string in red dots is a 6?pixel SRB string.Its reference pixels are the 15th SRB LUT pixel color duplicated six times.The 5th string in purple dots is a 5?pixel PRB string.Its reference string is locat?ed across the boundary between current CTU and left CTU. The PRB string and its reference string have exactly the same horizontal line shape and length of 5 pixels.

    In a string?matching encoder,the best matching reference string for a current starting pixel Pm,nbeing coded is found via the following steps:

    1)Searching the best reference PRB string for a current string curSm,nvia three sub?steps:

    a)Calculating the hash value ha_valm,nof Pm,n.

    b)For the pixel coordinates(i,j)obtained from the hash chain with the same hash value ha_valm,n,determining the lon?gest matching reference string refSi,j={Si,j(k):0≤k<Li,j},i.e.de?termining the largest Li,jthat satisfies the constraint|Si,j(k)Y?Sm,n(k)Y|≤E,|Si,j(k)U?Sm,n(k)U|≤E,|Si,j(k)V?Sm,n(k)V|≤E for all k<Li,j,where E is a predetermined threshold.After going through all the pixel coordinates(i,j)on the hash chain of hash value ha_valm,n,multiple matching reference strings are found as best reference PRB string candidates.

    c)The best reference PRB string is selected from the best reference PRB string candidates,based on average RD cost evaluation.For a given current string curSm,nand its reference string refSi,jof length L,the average RD cost is calculated by(1)

    where bits(refSi,j)is the number of bits for coding the reference string refSi,j,λ is a weighting factor,and distortion(curSm,n,refSi,j)is the distortion between the current string curSm,nand the refer?ence string refSi,j.

    2)Searching the best reference SRB string,which is straightfor?ward because a reference SRB string is a simple duplication of an SRB pixel color.

    3)Either the best reference PRB string or the best reference SRB string is selected as the best matching reference string based on average RD cost evaluation.

    If no PRB string or SRB string can be found for the current pixel Pm,nbeing coded,the pixel itself is coded directly as an unmatched pixel.

    5 Experimental Results

    As an implementation example,the proposed string?match?ing technique is integrated into HM?16.4+SCM?4.0 reference software[31].All experimental results were generated under the common test conditions and configurations defined in[30]for HEVC SCC project.

    Thirteen test sequences(Table 1)are used in the experi?ment.The test sequences are classified into four categories:text and graphics with motion(TGM),mixed content(MC),camera captured(CC),and animation(ANI).Each test se?quence has a RGB color format version and a YCbCr(YUV)color format version.Therefore,there are 26 sequences in total used in the experiment.

    To evaluate the overall coding performance,the HEVC BD?rate metric[32],[33]is used for lossy coding and bit?rate sav?ing metric is used for lossless coding.In lossy coding,an aver?age BD?rate reduction is calculated by three color components G/Y,B/U,and R/V for each color format and category.In loss?less coding,four bit?rate saving values(total,average,mini?mum,and maximum)are calculated for each color format and category.Both lossy coding and lossless coding experiments used three configurations:all intra(AI),random access(RA),and low delay B(LB).Encoding and decoding software runtime were also compared for evaluating the complexity of the encod?er and decoder.

    Two coding methods were compared:

    1)HM?16.4+SCM?4.0(SCM)with default setting.In particular,the IBC search range is full frame.

    2)HM?16.4+SCM?4.0 integrated with string?matching tech?nique(SCM?SM).The string?matching search range is the current LCU and left LCU.

    Table 2 shows coding performance comparison(BD?rate re?duction percentage in negative numbers)between SCM and SCM?SM in the lossy coding case.Table 3 shows coding per?formance comparison(bit?rate saving percentage in negative numbers)between SCM and SCM?SM in the lossless coding case.Each row of data in the two tables corresponds to a com?bination of one color format and one category.There are totally eight combinations.Each combination contains 1-7 sequenc?es.The encoding and decoding runtime ratios are also shown in the tables.

    ▼Table 1.Four?category test sequences

    ▼Table 2.Performance comparison between SCM and SCM?SM in the lossy coding case

    ▼Table 3.Performance comparison between SCM and SCM?SM in lossless coding case

    The experimental results include:

    1)In the lossy coding case,SCM?SM can achieve up to 4.33% for AI,2.68%for RA,2.04%for LB average BD?rate reduc?tion over SCM.

    2)In the lossless coding case,SCM?SM can achieve up to 14.5%for AI,9.9%for RA,6.46%for LB maximum bit?rate saving and 5.17%for AI,3.47%for RA,2.76%for LB aver?age bit?rate saving over SCM.

    3)The improvement of SCM?SM over SCM depends on the con?tents.In YUV TGM case,the average bit?rate saving is 5.17%in lossless coding case and the average BD?rate re?duction of components Y,U and V are 4.19%,4.31%and 4.33%,respectively in lossy coding case for AI configura?tion.SCM?SM also has good BD?rate reduction over SCM for mixed content in all configurations,but no coding perfor?mance improvement for camera captured and animation con?tents.

    4)In the lossy coding case,encoding runtime increase is about 22%for AI,11%for RA,7%for LB.In lossless coding case,encoding runtime increase is about 52%for AI,11% for RA,7%for LB

    6 Conclusions

    This paper proposes a string?matching coding technique for SCC.Both PRB and SRB are used for string?matching.The re?sulting bit-stream is a combination of PRB string coding param?eters,SRB string coding parameters,and unmatched pixelmixed on a string?by?string basis.The experiments using com?mon test condition[30]show that the string?matching coding technique can achieve a lossy coding BD?rate reduction of up to 4.33%,2.68%,2.04%for AI,RA,and LB configurations re?spectively,and a lossless coding average bit?rate saving of up to 5.17%,3.47%,2.76%for AI,RA,and LB configurations re?spectively.

    Our future work includes:1)implementing rate?control in string?matching enhanced coding system;2)optimizing string?matching coding to improve coding performance and reduce coding complexity further;3)achieving idempotent(re?loss?free)[35]coding in repetitive(multi?generation)compression using the string?matching enhanced coding system.

    References

    [1]T.Lin,K.Zhou,and S.Wang.“Cloudlet?screen computing:a client?server ar?chitecture with top graphics performance,”International Journal of Ad Hoc and Ubiquitous Computing,vol.13,no.2,pp.96-108,Jun.2013.doi:10.1504/IJA?HUC.2013.054174.

    [2]Y.Lu,S.Li,and H.Shen,“Virtualized screen:a third element for cloud_mobile convergence,”IEEE Multimedia,vol.18,no.2,pp.4-11,Apr.2011.doi:10.1109/MMUL.2011.33.

    [3]T.Lin and S.Wang.“Cloudlet?screen computing:a multi?core?based,cloud?computing?oriented,traditional?computing?compatible parallel computing par?adigm for the masses,”in IEEE International Conference on Multimedia and Ex?po,New York,USA,Jul.2009,pp.1805-1808.doi:10.1109/ICME.2009. 5202873.

    [4]T.Lin,K.Zhou,X.Chen,and S.Wang,“Arbitrary shape matching for screen content coding,”Picture Coding Symposium,San Jose,USA,Dec.2013,pp. 369-372.doi:10.1109/PCS.2013.6737760.

    [5]W.Zhu,J.Xu,W.Ding,et al.,“Adaptive LZMA?based coding for screen con?tent,”Picture Coding Symposium,San Jose,USA,Dec.2013,pp.373-376.doi:10.1109/PCS.2013.6737761.

    [6]T.Lin,X.Chen,and S.Wang,“Pseudo?2?D?matching based dual?coder archi?tecture for screen contents coding,”IEEE International Conference on Multime?dia and Expo Workshops,San Jose,USA,Jul.2013,pp.1-4.doi:10.1109/IC?MEW.2013.6618315.

    [7]S.Wang and T.Lin,“Compound image compression based on unified LZ and hy?brid coding,”IET Image Processing,vol.7,no.5,pp.484-499,May 2013.doi:10.1049/iet?ipr.2012.0439.

    [8]T.Lin,P.Zhang,S.Wang,et al.,“Mixed chroma sampling?rate high efficiency video coding for full?chroma screen content,”IEEE Transactions on Circuits and Systems for Video Technology,vol.23,no.1,pp.173-185,Jan.2013.doi:10.1109/TCSVT.2012.2223871.

    [9]W.Zhu,W.Ding,R.Xiong,et al.,“Compound image compression by multi?stage prediction,”IEEE Visual Communications and Image Processing Confer?ence,San Diego,USA,Nov.2012,pp.1-6.doi:10.1109/VCIP.2012.6410758.

    [10]S.Wang and T.Lin,“United coding method for compound image compres?sion,”Multimedia Tools and Applications,vol.71,no.3,pp.1263-1282,Aug. 2014.doi:10.1007/s11042?012?1274?y.

    [11]S.Wang and T.Lin,“United coding for compound image compression,”3rd In?ternational Congress on Image and Signal Processing,Yantai,China,pp.566-570,Oct.2010.doi:10.1109/CISP.2010.5647270.

    [12]S.Wang and T.Lin,“A unified LZ and hybrid coding for compound image par?tial?lossless compression,”2nd International Conference on Image and Signal Processing,Tianjin,China,Oct.2009,pp.1-5.doi:10.1109/CISP.2009. 5301019.

    [13]W.Ding,Y.Lu,and F.Wu,“Enable efficient compound image compression in h.264/AVC intra coding,”IEEE International Conference on Image Processing,San Antonio,USA,Oct.2007,vol.2,pp.337-340.doi:10.1109/ICIP.2007. 4379161.

    [14]ISO/IEC JTC1/SC29/WG11 and ITU?T Q6/16,“Joint call for proposals for cod?ing of screen content,”ISO/IEC JTC 1/SC 29/WG 11(MPEG),San Jose,USA,Doc.N14175,Jan.2014.

    [15]D.?K.Kwon and M.Budagavi,“Intra motion compensation with variable length intra MV coding,”JCT?VC,Vienna,Austria,JCTVC?N0206,Jul.2013.

    [16]C.Pang,J.Sole,L.Guo,et al.,“Intra motion compensation with 2?D MVs,”JCT?VC,Vienna,Austria,JCTVC?N0256,Jul.2013.

    [17]C.Pang,J.Sole,L.Guo,et al.,“Displacement vector signaling for intra block copying,”JCT?VC,Geneva,Switzerland,JCTVC?O0154,Oct.2013.

    [18]T.Lin,S.Wang,P.Zhang,and K.Zhou,“AHG7:full?chroma(YUV444)dic?tionary+hybrid dual?coder extension of HEVC,”JCT?VC,Shanghai,China,JCTVC?K0133,Oct.2012.

    [19]M.Budagavi,“Cross?check of JCTVC?K0133(dictionary+hybrid dual?coder extension of HEVC),”JCT?VC,Shanghai,China,JCTVC?K0329,Oct.2012.

    [20]T.Lin,S.Wang,P.Zhang,and K.Zhou,“AHG8:P2M based dual?coder exten?sion of HEVC,”JCT?VC,Geneva,Switzerland,JCTVC?L0303,Jan.2013.

    [21]J.Ye,S.Liu,S.Lei,et al.,“Improvements on 1D dictionary coding,”JCT?VC,Valencia,Spain,JCTVC?Q0124,2014.

    [22]R.Cohen,“Crosscheck for JCTVC?Q0124 improvements on 1D dictionary cod?ing mode,”JCT?VC,Valencia,Spain,JCTVC?Q0125,2014.

    [23]L.Zhao,X.Chen,T.Lin,et al.,“SCCE4:Results of subtest 3.3,”JCT?VC,Sapporo,Japan,JCTVC?R0060,2014.

    [24]B.Li and J.Xu,“Cross?check of test 3.3(JCTVC?R0060),”JCT?VC,Sappo?ro,Japan,JCTVC?R0109,2014.

    [25]K.Zhou,L.Zhao,X.Chen,and T.Lin,“Non?CE10:improvement on coding of ISC parameters and comparison to Palette coding,”JCT?VC,Strasbourg,F(xiàn)rance,JCTVC?S0250,Oct.2014.

    [26]L.Zhao,K.Zhou,and T.Lin,“CE3:results of test B.4.2(minimum string length of 20 pixels)on intra string copy,”JCT?VC,Geneva,Switzerland,JCT?VC?T0136,F(xiàn)eb.2015.

    [27]C.?H.Hung,Y.?J.Chang,J.?S.Tu,et al.,“CE3:crosscheck of CE3 test B.4.2(JCTVC?T0136),”JCT?VC,Geneva,Switzerland,JCTVC?T0179,F(xiàn)eb.2015.

    [28]L.Zhao,K.Zhou,S.Wang,and T.Lin,“Non?CE3:improvement on intra string copy,”JCT?VC,Geneva,Switzerland,JCTVC?T0139,F(xiàn)eb.2015.

    [29]R.?L.Liao,C.?C.Chen,W.?H.Peng,and H.?M.Hang,“Crosscheck of non?CE3:improvement on intra string copy(JCTVC?T0139),”JCT?VC,Geneva,Switzerland,JCTVC?T0200,F(xiàn)eb.2015.

    [30]H.Yu,R.Cohen,K.Rapaka,and J.Xu,“Common conditions for screen con?tent coding tests,”JCT?VC,Strasbourg,F(xiàn)rance,JCTVC?S1015,Oct.2014.

    [31]Heinrich Hertz Institute.(2015).Rec.ITU?T H.265|ISO/IEC 23008?2 High Efficiency Video Coding[Online].Available:https://hevc.hhi.fraunhofer.de/svn/ svn_HEVCSoftware/tags/HM?16.4+SCM?4.0

    [32]G.Bj?ntegaard,“Calculation of average PSNR differences between RD?Curves,”ITU?T SG16 Q.6 Document,VCEG?M33,Austin,USA,Apr.2001.

    [33]G.Bj?ntegaard,“Improvements of the BD?PSNR model,”ITU?T SG16 Q.6 Document,VCEG?AI11,Berlin,Germany,Jul.2008.

    [34]T.V.Ramabadran and S.S.Gaitonde,“A tutorial on CRC computations,”IEEE Micro,vol.8,no.4,pp.62-75,Aug.1988.doi:10.1109/40.7773.

    [35]T.Lin,“Achieving re?loss?free video coding,”IEEE Signal Processing Letters,vol.16,no.4,pp.323-326,Apr.2009.doi:10.1109/LSP.2009.2014285.

    Manuscript received:2015?05?13

    Biographies

    Tao Lino Lin(lintao@#edu.cn)received his MS and PhD degrees from Tohoku Uni?versity,Japan,in 1985 and 1989.He has been with VLSI Lab,Tongji University,China since 2003.In 2005,he was awarded“Chang Jiang Scholars”,the highest honor given by China Ministry of Education.From 1988 to 2002,he was a postdoc?toral researcher with University of California,Berkeley,and developed multimedia ICs and products at several companies in Silicon Valley,including Integrated De?vice Technology,Inc.,PMC?Sierra Inc.,Cypress Semiconductor Corp.,and NeoMag?ic Corp.He has been granted 24 US patents and 14 China patents.His current re?search interests include cloud?mobile computing,digital signal processing,audiovi?sual coding,and multimedia SoC design.

    Kailun Zhou Zhou(vlsi@#edu.cn)received his MS degree from Shanghai Jiaotong University,China in 2003.He is currently pursuing the PhD degree with Tongji Uni?versity,China.His current research interests include embedded system design,vid?eo coding,and ASIC architecture,design and verification.

    Liping Zhao Zhao(vlsi@#edu.cn)received her MS degree in computer science and technology from Hunan University,China in 2009.She is currently a PhD candidate in control science and engineering at VLSI lab of Tongji University,China.Her cur?rent research interests include screen content coding and video coding.

    This work was supported in part by National Natural Science Foundation of China under Grant No.61201226 and 61271096,Natural Science Foundation of Shanghai under Grant No.12ZR1433800,and Specialized Research Fund for the Doctoral Program under Grant No.20130072110054.

    av天堂久久9| 日本免费一区二区三区高清不卡 | xxx96com| 精品熟女少妇八av免费久了| 国产精品爽爽va在线观看网站 | 亚洲国产欧美日韩在线播放| 亚洲三区欧美一区| 免费在线观看日本一区| 亚洲免费av在线视频| 国产精品久久久久久人妻精品电影| 岛国视频午夜一区免费看| 免费在线观看视频国产中文字幕亚洲| 久久久久国产一级毛片高清牌| 电影成人av| 日韩视频一区二区在线观看| 在线观看一区二区三区| 国产精品一区二区精品视频观看| 国产人伦9x9x在线观看| 露出奶头的视频| 日韩 欧美 亚洲 中文字幕| 麻豆久久精品国产亚洲av | 欧美av亚洲av综合av国产av| 999久久久精品免费观看国产| 国产男靠女视频免费网站| 亚洲中文av在线| 国产精品亚洲一级av第二区| 久久久国产一区二区| 黄片大片在线免费观看| 午夜免费观看网址| 日本五十路高清| 久久香蕉精品热| 超碰97精品在线观看| 天堂俺去俺来也www色官网| 一边摸一边抽搐一进一出视频| 搡老乐熟女国产| 久久草成人影院| 午夜a级毛片| 亚洲自偷自拍图片 自拍| 黑人猛操日本美女一级片| 国产亚洲精品久久久久5区| 国产成+人综合+亚洲专区| 又大又爽又粗| 国产精品久久久人人做人人爽| 涩涩av久久男人的天堂| 一进一出抽搐gif免费好疼 | 国产伦一二天堂av在线观看| 怎么达到女性高潮| 人妻久久中文字幕网| 亚洲人成伊人成综合网2020| 日韩大码丰满熟妇| 欧美激情极品国产一区二区三区| 亚洲视频免费观看视频| 日韩精品青青久久久久久| 国产免费男女视频| 最新在线观看一区二区三区| 一区二区日韩欧美中文字幕| 亚洲一区中文字幕在线| 女人被狂操c到高潮| 欧美精品一区二区免费开放| 中文字幕人妻丝袜一区二区| 日韩欧美一区视频在线观看| 丁香六月欧美| 亚洲男人的天堂狠狠| 自拍欧美九色日韩亚洲蝌蚪91| 久久久精品国产亚洲av高清涩受| 久久久久久久精品吃奶| 欧美日本中文国产一区发布| 十八禁网站免费在线| 日韩免费av在线播放| 亚洲人成网站在线播放欧美日韩| 日韩精品中文字幕看吧| 精品卡一卡二卡四卡免费| 日日爽夜夜爽网站| 国产精品永久免费网站| 99久久精品国产亚洲精品| 久久人妻福利社区极品人妻图片| 丰满的人妻完整版| 亚洲片人在线观看| 日本免费一区二区三区高清不卡 | 黄色毛片三级朝国网站| 日本一区二区免费在线视频| 午夜精品久久久久久毛片777| 国产有黄有色有爽视频| 久久中文字幕一级| 国产一区二区在线av高清观看| 亚洲成人国产一区在线观看| 午夜免费成人在线视频| 人人妻,人人澡人人爽秒播| 色哟哟哟哟哟哟| 9热在线视频观看99| 岛国在线观看网站| 国产激情久久老熟女| 女警被强在线播放| 欧美乱妇无乱码| 亚洲国产欧美一区二区综合| 国产一区二区在线av高清观看| 美女午夜性视频免费| 国产欧美日韩一区二区三区在线| 夜夜看夜夜爽夜夜摸 | 好男人电影高清在线观看| 亚洲国产欧美日韩在线播放| 又紧又爽又黄一区二区| 午夜激情av网站| 欧美激情久久久久久爽电影 | www.精华液| 自拍欧美九色日韩亚洲蝌蚪91| 亚洲片人在线观看| 少妇的丰满在线观看| 很黄的视频免费| 69精品国产乱码久久久| 久久久久国内视频| 亚洲欧美一区二区三区黑人| 欧美在线一区亚洲| 亚洲欧美一区二区三区黑人| 久久久国产精品麻豆| 人人妻人人爽人人添夜夜欢视频| 亚洲国产看品久久| av天堂在线播放| 国产高清视频在线播放一区| 一区二区三区激情视频| 一区二区三区激情视频| 精品福利观看| 欧美不卡视频在线免费观看 | 中出人妻视频一区二区| 在线av久久热| av片东京热男人的天堂| 中文欧美无线码| 国产精品 国内视频| 国产有黄有色有爽视频| 免费女性裸体啪啪无遮挡网站| 久久久久久久午夜电影 | 国产97色在线日韩免费| av网站在线播放免费| 脱女人内裤的视频| 国产成+人综合+亚洲专区| 亚洲一区二区三区不卡视频| 欧美成人午夜精品| 久久欧美精品欧美久久欧美| 天天躁狠狠躁夜夜躁狠狠躁| av中文乱码字幕在线| 免费看a级黄色片| 色在线成人网| 精品人妻1区二区| 亚洲自拍偷在线| 美女午夜性视频免费| 一a级毛片在线观看| 老司机深夜福利视频在线观看| 99久久国产精品久久久| 成人黄色视频免费在线看| 色精品久久人妻99蜜桃| 中文欧美无线码| 亚洲成a人片在线一区二区| 日韩有码中文字幕| 真人一进一出gif抽搐免费| 久久九九热精品免费| 午夜精品在线福利| 国产色视频综合| 香蕉国产在线看| 久久香蕉激情| aaaaa片日本免费| 日韩欧美在线二视频| 免费在线观看日本一区| av电影中文网址| 欧美不卡视频在线免费观看 | 正在播放国产对白刺激| 亚洲情色 制服丝袜| 50天的宝宝边吃奶边哭怎么回事| 看黄色毛片网站| 欧美人与性动交α欧美软件| 亚洲午夜精品一区,二区,三区| 热99国产精品久久久久久7| 久热爱精品视频在线9| 亚洲欧美精品综合一区二区三区| 又紧又爽又黄一区二区| 国产精品1区2区在线观看.| 在线观看www视频免费| 午夜精品国产一区二区电影| 亚洲美女黄片视频| 一级a爱片免费观看的视频| 日本免费一区二区三区高清不卡 | 国产伦人伦偷精品视频| 日韩欧美一区视频在线观看| 午夜精品国产一区二区电影| 婷婷精品国产亚洲av在线| 国产三级在线视频| 97碰自拍视频| 日韩欧美一区视频在线观看| 在线永久观看黄色视频| 国产精品影院久久| 日本撒尿小便嘘嘘汇集6| 黄色视频不卡| 久久久久久久午夜电影 | 亚洲激情在线av| 天堂动漫精品| 俄罗斯特黄特色一大片| 99riav亚洲国产免费| 久久久精品国产亚洲av高清涩受| 搡老乐熟女国产| 免费日韩欧美在线观看| 亚洲成人久久性| 精品国产一区二区三区四区第35| 人人妻人人爽人人添夜夜欢视频| 日韩欧美免费精品| 国产黄色免费在线视频| 亚洲成a人片在线一区二区| 亚洲在线自拍视频| 欧美 亚洲 国产 日韩一| 黑人巨大精品欧美一区二区mp4| 午夜免费成人在线视频| 精品国产美女av久久久久小说| 亚洲av美国av| 精品国产乱码久久久久久男人| 国产色视频综合| 自拍欧美九色日韩亚洲蝌蚪91| 丁香六月欧美| 国产视频一区二区在线看| 国产野战对白在线观看| 男女床上黄色一级片免费看| 人成视频在线观看免费观看| 无人区码免费观看不卡| 91大片在线观看| www日本在线高清视频| 国产视频一区二区在线看| 亚洲国产毛片av蜜桃av| 美女高潮到喷水免费观看| 亚洲一区二区三区欧美精品| 日本黄色日本黄色录像| 露出奶头的视频| 日韩av在线大香蕉| 色在线成人网| 韩国精品一区二区三区| 一级a爱视频在线免费观看| 99久久久亚洲精品蜜臀av| 欧美成狂野欧美在线观看| 国产亚洲精品综合一区在线观看 | 午夜影院日韩av| 一级黄色大片毛片| 身体一侧抽搐| 久久国产乱子伦精品免费另类| 五月开心婷婷网| 中文字幕另类日韩欧美亚洲嫩草| 亚洲色图 男人天堂 中文字幕| 高清在线国产一区| 91在线观看av| 999精品在线视频| av欧美777| 久久久久久久午夜电影 | 日韩大尺度精品在线看网址 | 国产亚洲精品一区二区www| 一级黄色大片毛片| 国产男靠女视频免费网站| 88av欧美| 欧美成人性av电影在线观看| 一区在线观看完整版| 欧洲精品卡2卡3卡4卡5卡区| 亚洲午夜理论影院| 一进一出抽搐gif免费好疼 | 精品久久久久久电影网| 黄片播放在线免费| 黑人操中国人逼视频| 亚洲欧美激情在线| 91字幕亚洲| 亚洲激情在线av| 俄罗斯特黄特色一大片| 欧美成狂野欧美在线观看| 亚洲男人的天堂狠狠| 咕卡用的链子| 久久香蕉精品热| 久久久国产欧美日韩av| 老汉色av国产亚洲站长工具| a级毛片在线看网站| 中文字幕人妻熟女乱码| 国产成人欧美在线观看| 一级片'在线观看视频| 久久伊人香网站| 免费av中文字幕在线| 看免费av毛片| 国产亚洲精品一区二区www| 欧美国产精品va在线观看不卡| 欧美激情久久久久久爽电影 | 亚洲 欧美一区二区三区| 国产精品亚洲一级av第二区| 在线观看一区二区三区激情| 黄色视频不卡| 在线观看日韩欧美| 久久久久国产精品人妻aⅴ院| 在线免费观看的www视频| 黄色片一级片一级黄色片| 老熟妇乱子伦视频在线观看| 久久精品国产清高在天天线| 久久这里只有精品19| 国产免费男女视频| a级片在线免费高清观看视频| 高清毛片免费观看视频网站 | 亚洲中文日韩欧美视频| 免费少妇av软件| 久久久久国内视频| 啦啦啦在线免费观看视频4| 亚洲精品国产区一区二| 80岁老熟妇乱子伦牲交| 99精品久久久久人妻精品| 亚洲专区字幕在线| 村上凉子中文字幕在线| 超色免费av| 日日爽夜夜爽网站| 亚洲精品在线观看二区| 男女之事视频高清在线观看| 自拍欧美九色日韩亚洲蝌蚪91| 曰老女人黄片| 亚洲久久久国产精品| 后天国语完整版免费观看| 看免费av毛片| 久久久久亚洲av毛片大全| 久久久久久久午夜电影 | 人妻久久中文字幕网| 一级毛片女人18水好多| 激情在线观看视频在线高清| 巨乳人妻的诱惑在线观看| 大陆偷拍与自拍| 国产国语露脸激情在线看| 久久影院123| 高清欧美精品videossex| 97碰自拍视频| 久久久水蜜桃国产精品网| 交换朋友夫妻互换小说| 欧美中文日本在线观看视频| 窝窝影院91人妻| 欧美日韩瑟瑟在线播放| 99久久人妻综合| 黑丝袜美女国产一区| 成熟少妇高潮喷水视频| 色综合婷婷激情| 色哟哟哟哟哟哟| 亚洲av成人av| 两性夫妻黄色片| 91大片在线观看| 国产成+人综合+亚洲专区| 亚洲 国产 在线| 亚洲第一青青草原| 亚洲黑人精品在线| 老司机午夜十八禁免费视频| 制服人妻中文乱码| 亚洲午夜精品一区,二区,三区| 午夜影院日韩av| 嫩草影院精品99| 一级作爱视频免费观看| 黄片播放在线免费| 久久国产精品影院| 欧美在线黄色| 中文字幕人妻丝袜制服| 老司机福利观看| 一级毛片精品| 真人做人爱边吃奶动态| 亚洲中文字幕日韩| 亚洲成人精品中文字幕电影 | 亚洲九九香蕉| 无人区码免费观看不卡| 99国产精品99久久久久| 久久久久久人人人人人| 欧美黑人欧美精品刺激| 国产精华一区二区三区| www.熟女人妻精品国产| 亚洲专区字幕在线| 手机成人av网站| 在线视频色国产色| 岛国视频午夜一区免费看| 久久天躁狠狠躁夜夜2o2o| 免费久久久久久久精品成人欧美视频| 国产成人精品久久二区二区免费| 亚洲一区二区三区色噜噜 | 成人免费观看视频高清| 免费在线观看黄色视频的| 丁香六月欧美| 老熟妇仑乱视频hdxx| 日韩三级视频一区二区三区| 超色免费av| 国产成人av激情在线播放| 不卡一级毛片| 日本精品一区二区三区蜜桃| 国产黄色免费在线视频| svipshipincom国产片| 午夜激情av网站| 日韩 欧美 亚洲 中文字幕| 男女之事视频高清在线观看| 国产亚洲欧美98| 亚洲自偷自拍图片 自拍| 天天影视国产精品| 一个人观看的视频www高清免费观看 | 欧美乱码精品一区二区三区| 国产一区二区三区视频了| 亚洲久久久国产精品| 亚洲色图 男人天堂 中文字幕| 人妻丰满熟妇av一区二区三区| 老司机靠b影院| 19禁男女啪啪无遮挡网站| 国产色视频综合| 日本wwww免费看| 久久国产亚洲av麻豆专区| 岛国在线观看网站| 新久久久久国产一级毛片| 亚洲九九香蕉| 成熟少妇高潮喷水视频| 变态另类成人亚洲欧美熟女 | 国产av在哪里看| 亚洲av电影在线进入| 久久久国产一区二区| 又紧又爽又黄一区二区| 国产av一区二区精品久久| 嫁个100分男人电影在线观看| 我的亚洲天堂| 999久久久国产精品视频| 国产一区二区在线av高清观看| 亚洲人成伊人成综合网2020| 久久天堂一区二区三区四区| 母亲3免费完整高清在线观看| 中文字幕最新亚洲高清| 亚洲色图av天堂| 另类亚洲欧美激情| 欧美成人性av电影在线观看| 欧洲精品卡2卡3卡4卡5卡区| 国产精品亚洲一级av第二区| 天堂动漫精品| 99香蕉大伊视频| 91在线观看av| 在线观看一区二区三区| 又大又爽又粗| 看黄色毛片网站| 无遮挡黄片免费观看| 电影成人av| 淫妇啪啪啪对白视频| 伊人久久大香线蕉亚洲五| 欧美日韩乱码在线| 亚洲国产中文字幕在线视频| 欧美成人午夜精品| 国产一区二区三区综合在线观看| 99热国产这里只有精品6| 在线观看舔阴道视频| 亚洲国产看品久久| 久久精品成人免费网站| xxxhd国产人妻xxx| 亚洲美女黄片视频| 日韩人妻精品一区2区三区| 久久影院123| 大型黄色视频在线免费观看| 香蕉丝袜av| 亚洲欧洲精品一区二区精品久久久| ponron亚洲| 国产成人精品无人区| 黄色a级毛片大全视频| 国产亚洲精品第一综合不卡| 日韩av在线大香蕉| 人人妻人人添人人爽欧美一区卜| 另类亚洲欧美激情| 狂野欧美激情性xxxx| 每晚都被弄得嗷嗷叫到高潮| 电影成人av| 看片在线看免费视频| 久久精品aⅴ一区二区三区四区| 久久性视频一级片| 亚洲专区字幕在线| 亚洲熟妇中文字幕五十中出 | 18禁黄网站禁片午夜丰满| 亚洲avbb在线观看| 免费在线观看视频国产中文字幕亚洲| 日本 av在线| 欧美老熟妇乱子伦牲交| 免费在线观看黄色视频的| 一边摸一边抽搐一进一出视频| www.999成人在线观看| 在线免费观看的www视频| 欧美激情极品国产一区二区三区| 国产亚洲欧美98| 日韩中文字幕欧美一区二区| 黄色视频,在线免费观看| 高清在线国产一区| 精品人妻1区二区| 久久久久九九精品影院| 日日爽夜夜爽网站| 日韩 欧美 亚洲 中文字幕| 欧美色视频一区免费| 欧美国产精品va在线观看不卡| 国产成年人精品一区二区 | 国产精品电影一区二区三区| av中文乱码字幕在线| 日韩大码丰满熟妇| 精品国产国语对白av| 男女床上黄色一级片免费看| 国产精品美女特级片免费视频播放器 | 99精国产麻豆久久婷婷| 久久影院123| 丁香欧美五月| 欧美日韩视频精品一区| 国产精品久久久久成人av| 国产精品二区激情视频| 一级片'在线观看视频| 韩国精品一区二区三区| 欧美色视频一区免费| 亚洲精品在线美女| 日韩av在线大香蕉| 在线国产一区二区在线| 一进一出好大好爽视频| 亚洲第一欧美日韩一区二区三区| 999久久久国产精品视频| 麻豆av在线久日| 久久精品国产综合久久久| 岛国视频午夜一区免费看| 久久草成人影院| 婷婷精品国产亚洲av在线| 中文欧美无线码| 亚洲国产精品合色在线| 国产精品爽爽va在线观看网站 | 9热在线视频观看99| 国产亚洲欧美在线一区二区| 夫妻午夜视频| 国产xxxxx性猛交| 精品国产一区二区久久| 午夜两性在线视频| 69av精品久久久久久| 一级毛片女人18水好多| 精品国产美女av久久久久小说| 成人国语在线视频| 午夜视频精品福利| 99riav亚洲国产免费| 久久久久久亚洲精品国产蜜桃av| 成年版毛片免费区| 极品教师在线免费播放| 一区二区三区国产精品乱码| 国产精品九九99| 欧美大码av| 午夜a级毛片| 精品一区二区三区四区五区乱码| 亚洲成人免费电影在线观看| 日本免费a在线| 国产成人免费无遮挡视频| 亚洲成人国产一区在线观看| 亚洲第一欧美日韩一区二区三区| 女人被躁到高潮嗷嗷叫费观| 麻豆av在线久日| 免费在线观看视频国产中文字幕亚洲| 国产成人精品在线电影| 久久久久久久久免费视频了| 久热这里只有精品99| 老司机福利观看| 亚洲自拍偷在线| 日本vs欧美在线观看视频| 国产一区二区三区视频了| 99riav亚洲国产免费| 九色亚洲精品在线播放| 国产三级黄色录像| 最新美女视频免费是黄的| 国产免费av片在线观看野外av| 久久久国产一区二区| 波多野结衣av一区二区av| 久久这里只有精品19| 大型黄色视频在线免费观看| 别揉我奶头~嗯~啊~动态视频| 在线观看www视频免费| 久久久久九九精品影院| 水蜜桃什么品种好| 中亚洲国语对白在线视频| 一边摸一边做爽爽视频免费| 亚洲五月婷婷丁香| 国产欧美日韩一区二区三区在线| 97超级碰碰碰精品色视频在线观看| 免费看a级黄色片| 成年女人毛片免费观看观看9| 亚洲国产精品999在线| 老司机午夜福利在线观看视频| 国产成人欧美| 中文字幕另类日韩欧美亚洲嫩草| 看黄色毛片网站| 免费高清在线观看日韩| 在线观看舔阴道视频| 精品免费久久久久久久清纯| 久久精品亚洲精品国产色婷小说| 90打野战视频偷拍视频| 香蕉久久夜色| 91av网站免费观看| 欧美精品啪啪一区二区三区| 亚洲欧美一区二区三区黑人| 好看av亚洲va欧美ⅴa在| 午夜福利在线免费观看网站| 国产成人精品久久二区二区免费| 国产一区二区三区综合在线观看| 精品一区二区三区av网在线观看| 欧美老熟妇乱子伦牲交| 级片在线观看| 啦啦啦免费观看视频1| 久久草成人影院| 成人三级黄色视频| 首页视频小说图片口味搜索| 亚洲成人精品中文字幕电影 | 91麻豆av在线| 国产一区在线观看成人免费| 精品久久蜜臀av无| 天天影视国产精品| 国内久久婷婷六月综合欲色啪| 男女做爰动态图高潮gif福利片 | a级毛片在线看网站| 亚洲午夜精品一区,二区,三区| av超薄肉色丝袜交足视频| 在线观看免费视频日本深夜| 一边摸一边抽搐一进一小说| 久久国产精品影院| 欧美精品一区二区免费开放| 一进一出抽搐gif免费好疼 | 亚洲中文日韩欧美视频| 成年人免费黄色播放视频| 黄色毛片三级朝国网站| 老汉色av国产亚洲站长工具| 丁香六月欧美| 五月开心婷婷网| 91在线观看av| 国内毛片毛片毛片毛片毛片| 日韩免费高清中文字幕av|