서지주요정보
Blind post-processing for compression artifact reduction in low bit-rate coded videos = 저 전송률 부호화 동영상의 압축 아티팩트 감소를 위한 블라인드 후처리 기법
서명 / 저자 Blind post-processing for compression artifact reduction in low bit-rate coded videos = 저 전송률 부호화 동영상의 압축 아티팩트 감소를 위한 블라인드 후처리 기법 / Seok Bong Yoo.
발행사항 [대전 : 한국과학기술원, 2015].
Online Access 원문보기 원문인쇄

소장정보

등록번호

8028515

소장위치/청구기호

학술문화관(문화관)B1층 보존서고

DEE 15058

휴대폰 전송

도서상태

이용가능(대출불가)

사유안내

반납예정일

리뷰정보

초록정보

Recent video coding standards produce unwanted spatial and temporal artifacts in reconstructed videos. In this dissertation, the reduction of major coding artifacts, such as ringing and blocking artifacts, is considered. Ringing or mosquito artifacts arise due to the quantization of high frequency discrete cosine transform coefficients. Most of the existing artifact reduction algorithms assume that coding information such as a standard quantization table and the corresponding quantization parameter for each block are available. In many multimedia applications, however, external video inputs are usually supplied without coding information. To effectively reduce the ringing and mosquito artifacts in a decoded input video sequence, it is necessary to control the filter strength block by block. In this dissertation, we present a blind block-based algorithm to estimate the quantization amount and propose a novel post-processing algorithm for ringing and mosquito artifact reduction based on the visibility of artifacts in terms of the human visual system using the estimated quantization amount. Meanwhile, block-based coding introduces an undesirable discontinuity between neighboring blocks in reconstructed images. This image degradation, referred to as blocking artifacts, arises mainly due to the loss of inter-block correlation in the quantization process of discrete cosine transform coefficients. In this dissertation, we presents a novel post-processing algorithm based on increment of inter-block correlation aimed at reducing blocking artifacts. We first smooth the three lowest frequency discrete cosine transform coefficients between neighboring blocks, in order to reduce blocking artifacts in the flat region, which are most sensitive to the human visual system. We then group each edge block and its matched blocks together and apply group-based filtering to increase the correlation between grouped blocks. This suppresses blocking artifacts in the edge region while preserving details. In addition, the algorithm is extended to reduce flickering artifacts as well as blocking artifacts in video sequences. Experimental results show that proposed deringing and deblocking algorithms successfully alleviate ringing and blocking artifacts, respectively, in videos coded with low bit-rates, regardless of video contents and resolutions.

영상 데이터는 그 크기가 매우 크기 때문에 영상의 전송, 저장을 위해서 필연적으로 영상의 부호화 과정이 요구된다. 하지만 저 전송률 환경의 영상의 부호화는 재구성된 영상에서 원치 않는 압축 아티팩트를 생성한다. 예를 들면, 고주파 이산 변환 계수의 양자화 에러로 인하여 재구성된 영상의 강한 에지 영역에서 ringing 잡음이 발생된다. 동영상의 경우 이러한 ringing 잡음의 시간 축 변동으로 인해 시간 축으로의 mosquito 잡음이 함께 발생된다. 본 학위 논문의 첫 번째 파트에서는 이러한 ringing 및 mosquito 잡음과 같은 주요 압축 아티팩트의 감소를 위한 후처리 알고리즘을 제안한다. 기존의 압축 아티팩트 감소 알고리즘의 대부분은 표준 양자화 테이블과 각 블록에 해당하는 양자화 파라미터 등의 코딩 정보를 알고 있다고 가정한다. 그러나 외부 비디오 입력이 코딩 정보 없이 공급되는 상황이 다수 존재한다. 이 경우에 효과적으로 압축 아티팩트를 줄이기 위해서는 코딩 정보의 추정을 통한 압축 아티팩트 감소 알고리즘이 필요하다. 본 학위 논문에서는 코딩 정보가 없는 블라인드 환경에서 블록 기반으로 양자화 정도를 추정하고, 시각 시스템의 관점에서의 아티팩트의 시인성에 기초한 ringing과 mosquito 잡음 제거 알고리즘을 제시하였다. 한편, 블록 기반 부호화를 이용하여 재구성 된 영상에서는 블록 경계 부근에서 눈에 거슬리는 불연속을 추가로 발생시킨다. 이러한 블록 현상은 이산 코사인 변환의 양자화 과정에서 블록 간 상관 관계의 손실로 인해 발생되고, 본 학위 논문의 두 번째 파트에서는 이 블록 현상을 줄이기 위한 후처리 알고리즘을 제안한다. 제안하는 알고리즘에서는 시각 시스템에 가장 민감한 평탄 영역 내 블록 현상을 감소시키기 위해, 먼저 이웃한 블록들 사이의 세 개의 최저 주파수 변환 계수의 상관성을 증가시킨다. 이후에 블록 매칭을 통한 에지 블록들의 그룹화를 수행하고, 그룹화 된 에지 블록간의 상관 관계를 높이기 위해 적응적인 그룹 기반 필터링을 적용함으로써 에지 영역 내 블록 현상을 감소시킨다. 추가적으로, 동영상의 경우는 평탄 영역의 flickering 현상을 함께 줄이기 위해 시간 축으로 확장된 알고리즘을 제안하였다. 실험 결과, 제안한 deringing 및 deblocking 알고리즘이 영상 콘텐츠와 해상도에 관계없이 저 전송률로 압축된 동영상에 존재하는 아티팩트들을 성공적으로 감소시키는 것을 확인하였다. 또한 기존 최신 논문의 알고리즘과의 정성적, 정량적 결과 비교를 통하여 제안하는 deringing 및 deblocking 알고리즘의 성능의 우수성을 검증하였다.

서지기타정보

서지기타정보
청구기호 {DEE 15058
형태사항 x, 98p : 삽화 ; 30 cm
언어 영어
일반주기 저자명의 한글표기 : 유석봉
지도교수의 영문표기 : Jong Beom Ra
지도교수의 한글표기 : 나종범
수록잡지명 : "Blind post-processing for ringing and mosquito artifact reduction in coded videos". IEEE Transactions on Circuits and Systems for Video Technology, v.24, no.5, pp.721-732(2014)
수록잡지명 : "Post-processing for blocking artifact reduction based on inter-block correlation". IEEE Transactions on Multimedia, v.16, no.6, pp.1536-1548(2014)
Including Appendix
학위논문 학위논문(박사) - 한국과학기술원 : 전기및전자공학부,
서지주기 References : p.
QR CODE

책소개

전체보기

목차

전체보기

이 주제의 인기대출도서

In-loop deblockingfilteringrin theH.264/AVC encoder

Distributions of Euclidian distance values betweeninter-codedblocks andthecorresponding intra-quantized blocks for (a) two highbit-rates, (b) two medium bit-rates, and (c) two low bit-rates, respectively.

Various original blocks and thecorresponding coded ones usingMPEG-4andJPEG

Theproposed quantization amount estimation algorithm

(a) StandardJPEG quantizationtable, (b) standardMPEG quantizationtableforintra-mode.

Theproposed deringing and demosquito algorithm

(a) Complex and (b) less complex edgeregions. The upperrowis froman MJPEG-ooded Mobilevideoframe whilethelowerrowis fromaJPEG-ooded Cameramanimage.

Flowchart fordetermininga blocktype.

Less complexedgeblocks oredgeblocks belongingtoaless complexregion

Block type determination (a) An MJPEG-coded Mobile video frame and (b) the determinedblocktypes.

Variousblocks of type2andtheir coded ones usingMJPEGinthefirstandsecond rows. The third, fourth, fifth, sixth, and seventh rows show the deringing results obtained by using five differentschemes, respectively.

(a) Optimal values of03D,in whichprovidethehighestPSNRinblocks oftype1, and (b), (C) optimal values of O2D,i andOti. whichprovidethe highest PSNRinblocks oftype2. The values areplottedacoordingto thevalueof q.

Distributions of Euclidian distance values between inter-coded blocks and the conrespondingintra-quantizedblocksusingestimated quantization tables for(a) twohighbit-rates, (b) two mediumbit-rates, and(C) twolowbit-rates, respectively.

Various original blocks and the corresponding coded ones using MPEG-4 in the upper and middle rows. The bottom row shows coded blocks via intra-quantization using an estimated quantizationtable.

Accuracy oftheQP estimation for various intra-codedblocks

Comparison of different filtered versions of theMJPEG-codedresults of a videoframe, in terms ofPSNR.copp Thevideoframeis coded with differentbit-ratesina widerange.

Comparison of the differentfiltered versions of an MJPEG-coded Mobile sequence with a scalingfactorof4, interms of (a) PSNRedge and (b) flickeringmetric.

Comparison offilteringresults foran MJPEG-ooded Mobileframe. (a) Acoded frame with a scaling factor of 4 (PSNR-tige: 22.29 dB), and its filtered versions using (b) Liu's algorithm [18] (PSNRedge: 22.19dB), (C) Nath'salgorithm[15] (PSNRedge: 22.58 dB), (d) thespatialfuzzy filter[10] (PSNRetige: 22.52 dB), (e) MCSTF [12] (PSNRedge: 22.51 dB), and(f) proposedalgorithm (PSNRedge: 22.73dB), respectivel

Magrifiedlversions ofthefilteingresults mainlyincludingblocks oftype1inFig. 3.17.(a) A coded frameand its filtered versions using(b) Liu's algorithm [18], (C) Nath's algorithm[15], (d) thespatial fuzzyfilter[10], (e) MCSTF [12], and(f) proposedalgorithm, respectively.

Magnifiedversions offilteringrresults mainlyindudingblocks oftypes2and3inFig. 3.17. (a) Acodedframe and its filtered versions using(b) Liu'salgorithm[18], (c) Nath'salgorithm[15], (d) thespatial fuzzyfilter[10], (e) MCSTF[12], and(f) proposedalgorithm, respectively.

Comparison of the objective deringing quality without and with deblocking in terr PSNR(dB). The values in the parentheses denote the PSNRetge (dB) to emphasize the P improvementsintheedgeregion.

Comparison offilteringresults withoutdeblockingforan MJPEG-ooded Mobileframe. (a) A coded frame with a scale factor of4 (PSNR-tige: 22.29 dB), and its filtered versions using (b) the spatial fuzzy filter [10] (PSNRetgp: 22.50 dB), (c) MCSTF [12] (PSNRedge: 22.48 dB), and (d) proposedalgorithm(PSNRedge: 22.71 dB),respectively.

Processed result for an MJPEG-ooded Mobile colorframe. (a) Original, (b) a coded frame witha scalingfactorof4, and (C) its filtered version usingtheproposed algorithm

Comparison of the different filtered versions of an H.264-coded Foreman sequence in terms of(a) PSNRedgeand (b) flickeringmetric.

Comparison of the different filtered versions of an H.264-coded Highway sequence in terms of (a) PSNRedge and(b) flickering metric.

Comparison of filtering results for an H.264-ooded Foreman frame. (a) A coded frame (PSNRetpe: 28.21 dB) andits filtered versions using(b) Liu's algorithm[18] (PSNR-cige: 28.19dB), (C) Nath's algorithm[15] (PSNR-dg: 28.56dB), (d) thespatialfuzzy filter[10] (PSNRedige: 28.42dB), (e) MCSTF[12] (PSNRetg:: 28.61 dB), and(f) proposedalgorithm(PSNRedge: 28.81 dB).

Imagedecompositionby separatingDCT coefficientsinto two groups. (a), (d) Codedimages, (b), (e) LB images, and (C), (f) HBimages.

Coded image reprocessed by replacing the three LF DCT coefficients with their original values. (a), (d) Original images, (b), (e) codedimages, and(c), (f) reprocessedimages of (b) and(e).

Overall structure oftheproposed algorithm

Result ofblockclassification on (a) a codedimage and (b) its HB image. Edgeblocks are marked withwhiterectangies.

Asuperblockand thecorresponding secondaryblocks after8x8DCT

Quantization tables

Correlation distribution of (a) DC coefficients, (b) AC1 coefficients, and (C)AC2 coefficients

Refinement of three codedLF DCT coefficients.

Secondaryimages of (a) DC, (b) AC1, and (C) AC2 and (d)-(f) their filtered images. The imagesarere-ccaledfor enhanced display.

Result of flat region deblocking. (a) Coded image, (b) deblocked image based on the refinement ofthreeLFDCT coefficients.

Adaptive determination of thevalue of Cmatch acoordingito Oref.

Plots of O, whichprovidethehighestPSNRforfilteredimages, accordingtothevalueof 9

Anexampleof patch-basedboundary refinement.fora weakedgegroup.

Weighted stackingin theaggregationstageforedgeregiondeblocking

Aggregationresults inthe weakedgeregion. (a) Onginal, (b)JPEG withaqualityfactorof 20, (c) block-based 3D filtering[56], (d) proposed

Volumes consisting offiveconsecutivesecondaryimageframes of(a) DC, (b) AC1, and(c) AC2.

Plots of is, which provide the highest PSNR in flat region deblocking, according to the

Theparameter value of Hs aocordingito thevalueofQP

Average PSNR values of the proposed algorithm as a function of (a) Kand (b)Wpwan

Comparison of theobjective qualityin terms of PSNR (dB)

Comparison of theabjective qualityin terms of SSIM

Comparison of theobjective qualityin terms of PSNR-B (dB)

Comparison of the subjective visual quality forLena image (a) Onginal, (b) codedusing Q2, (c) Kim's, (d) BM3D filtering, (e) Foi's, (f) Sun's, (g) proposed.

Comparison of thesubjectivevisual quality forCameramanimage. (a) Original, (b) coded usingQ2, (C) Kim's, (d) Foi's, (e) Sun's, (f) proposed.

Comparisonofthesubjectivevisual qualityforBirdimage. (a) Onginal, (b) codedusingQ2, (c) Kim's, (d) Foi's, (e) Sun's, (f) proposed

Comparison of the subjective visual quality forLenaimage. (a) Original, (b) coded using Q1, (c) Kims, (d) Foi's, (e) Sun's, (f) proposed.

Comparison of the subjective visual quality for Barbara image. (a) Original, (b) coded usingQ2, (C) Kim's, (d) Foi's, (e) Sun's, (f) proposed.

Comparison of the subjective visual quality for various images. (a) Original, (b) coded usingQ3, (C) Kim's, (d) Foi's, (e) Sun's, (f) proposed.

Proposed deblocking result for a coded colorimage of Lena using Q2. (a) Oniginal, (b) coded, (c) proposed.

Deblocking results forthe H.264/AVC coded Hall monitor sequence (CIF, QP =42). (a) Coded frame withoutin-loop filtering, (b) coded frame with in-loop filtering, (c) the result of the proposedfilteringappliedto(a), and (d)-(f) theirmagnified versions.

Deblocking results for the H.265/HEVC coded Cactus sequence (Full HD, QP =42). (a) Coded withoutaay:inloopfiltering (b) coded withallin-loopfilterings, (C) theresultoftheproposed filteringappliedto (a).

Comparisonof theobjective qualityforvarious video sequences

Comparisonof theobjectivequalityforpost-processedvideo sequences codedv H.265/HEVC.

Post-processed results for H.265/HEVC coded BasketballDrive and BQTerrace sequences (FullHD, QP =42). (a) Coded withallin-loopfilterings, (b) theresults oftheRSTF appliedito(a), (C) theresults oftheproposedfilteringappliedto(a).

Post-processed results for a sequence downloaded from YouTube (4K UHD). (a)-(c) The 12th, 30th, and 62nd coded frames, (d)-(f) the results of the proposed filtering applied to (a)-(c), respectively.

Post-processed results for a sequence downloaded from YouTube (4K UHD). (a) Coded frame, (b) theresultoftheproposedfilteringappliedto(a).

Comparison of thesubjectivequalityin terms of mean opinionscores (MOS)

Comparison of thecomputational complexity of deblocking methods