A Hybrid DCT-SVD Video Compression Technique (HDCTSVD)



Similar documents
ON THE (Q, R) POLICY IN PRODUCTION-INVENTORY SYSTEMS

Questions & Answers Chapter 10 Software Reliability Prediction, Allocation and Demonstration Testing

AN IMPLEMENTATION OF BINARY AND FLOATING POINT CHROMOSOME REPRESENTATION IN GENETIC ALGORITHM

Software Engineering and Development

STUDENT RESPONSE TO ANNUITY FORMULA DERIVATION

Reduced Pattern Training Based on Task Decomposition Using Pattern Distributor

Effect of Contention Window on the Performance of IEEE WLANs

The transport performance evaluation system building of logistics enterprises

INITIAL MARGIN CALCULATION ON DERIVATIVE MARKETS OPTION VALUATION FORMULAS

HEALTHCARE INTEGRATION BASED ON CLOUD COMPUTING

Chapter 3 Savings, Present Value and Ricardian Equivalence

YARN PROPERTIES MEASUREMENT: AN OPTICAL APPROACH

Ilona V. Tregub, ScD., Professor

Tracking/Fusion and Deghosting with Doppler Frequency from Two Passive Acoustic Sensors

Cloud Service Reliability: Modeling and Analysis

An Epidemic Model of Mobile Phone Virus

Semipartial (Part) and Partial Correlation

An Approach to Optimized Resource Allocation for Cloud Simulation Platform

SUPPORT VECTOR MACHINE FOR BANDWIDTH ANALYSIS OF SLOTTED MICROSTRIP ANTENNA

An Introduction to Omega

CONCEPTUAL FRAMEWORK FOR DEVELOPING AND VERIFICATION OF ATTRIBUTION MODELS. ARITHMETIC ATTRIBUTION MODELS

Comparing Availability of Various Rack Power Redundancy Configurations

Research on Risk Assessment of the Transformer Based on Life Cycle Cost

Distributed Computing and Big Data: Hadoop and MapReduce

Comparing Availability of Various Rack Power Redundancy Configurations

A framework for the selection of enterprise resource planning (ERP) system based on fuzzy decision making methods

Automatic Closed Caption Detection and Filtering in MPEG Videos for Video Structuring

Strength Analysis and Optimization Design about the key parts of the Robot

The impact of migration on the provision. of UK public services (SRG ) Final Report. December 2011

STABILITY ANALYSIS IN MILLING BASED ON OPERATIONAL MODAL DATA 1. INTRODUCTION

Valuation of Floating Rate Bonds 1

Modal Characteristics study of CEM-1 Single-Layer Printed Circuit Board Using Experimental Modal Analysis

Physics 235 Chapter 5. Chapter 5 Gravitation

High Availability Replication Strategy for Deduplication Storage System

The Role of Gravity in Orbital Motion

Multiband Microstrip Patch Antenna for Microwave Applications

Vector Calculus: Are you ready? Vectors in 2D and 3D Space: Review

How To Use A Network On A Network With A Powerline (Lan) On A Pcode (Lan On Alan) (Lan For Acedo) (Moe) (Omo) On An Ipo) Or Ipo (

Uncertain Version Control in Open Collaborative Editing of Tree-Structured Documents

PAN STABILITY TESTING OF DC CIRCUITS USING VARIATIONAL METHODS XVIII - SPETO pod patronatem. Summary

Modeling and Verifying a Price Model for Congestion Control in Computer Networks Using PROMELA/SPIN

Episode 401: Newton s law of universal gravitation

Continuous Compounding and Annualization

Data Center Demand Response: Avoiding the Coincident Peak via Workload Shifting and Local Generation

Efficient Redundancy Techniques for Latency Reduction in Cloud Systems

Peer-to-Peer File Sharing Game using Correlated Equilibrium

Supplementary Material for EpiDiff

METHODOLOGICAL APPROACH TO STRATEGIC PERFORMANCE OPTIMIZATION

UNIT CIRCLE TRIGONOMETRY

Channel selection in e-commerce age: A strategic analysis of co-op advertising models

2 r2 θ = r2 t. (3.59) The equal area law is the statement that the term in parentheses,

Optimizing Content Retrieval Delay for LT-based Distributed Cloud Storage Systems

est using the formula I = Prt, where I is the interest earned, P is the principal, r is the interest rate, and t is the time in years.

Adaptive Queue Management with Restraint on Non-Responsive Flows

9:6.4 Sample Questions/Requests for Managing Underwriter Candidates

A Capacitated Commodity Trading Model with Market Power

Approximation Algorithms for Data Management in Networks

THE DISTRIBUTED LOCATION RESOLUTION PROBLEM AND ITS EFFICIENT SOLUTION

Improving Software Effort Estimation Using Neuro-Fuzzy Model with SEER-SEM

Towards Automatic Update of Access Control Policy

Skills Needed for Success in Calculus 1

Things to Remember. r Complete all of the sections on the Retirement Benefit Options form that apply to your request.

Financing Terms in the EOQ Model

Voltage ( = Electric Potential )

Supporting Efficient Top-k Queries in Type-Ahead Search

MATHEMATICAL SIMULATION OF MASS SPECTRUM

Chapter 2 Valiant Load-Balancing: Building Networks That Can Support All Traffic Matrices

A Two-Step Tabu Search Heuristic for Multi-Period Multi-Site Assignment Problem with Joint Requirement of Multiple Resource Types

Manual ultrasonic inspection of thin metal welds

Over-encryption: Management of Access Control Evolution on Outsourced Data

Database Management Systems

Automatic Testing of Neighbor Discovery Protocol Based on FSM and TTCN*

Risk Sensitive Portfolio Management With Cox-Ingersoll-Ross Interest Rates: the HJB Equation

883 Brochure A5 GENE ss vernis.indd 1-2

Power Monitoring and Control for Electric Home Appliances Based on Power Line Communication

Exam #1 Review Answers

Secure Smartcard-Based Fingerprint Authentication

The Detection of Obstacles Using Features by the Horizon View Camera

Towards Realizing a Low Cost and Highly Available Datacenter Power Infrastructure

MULTIPLE SOLUTIONS OF THE PRESCRIBED MEAN CURVATURE EQUATION

INVESTIGATION OF FLOW INSIDE AN AXIAL-FLOW PUMP OF GV IMP TYPE

How To Find The Optimal Stategy For Buying Life Insuance

Transcription:

A Hybid - Video Compession echnique (H) ong, Lin; Rao, K.R. Abstact A new hybid - (H) video compession technique is poposed in this pape. Discete cosine tansfom () is widely used in video coding due to its high enegy compaction and efficient computation complexity, singula value decomposition () is a tansfom that povides optimal enegy compaction fo any data. and ae combined to achieve optimal pefomance of the tansfom pat in the poposed video compession technique. is used only fo the blocks fo which cannot povide good compession. he decision citeion is set in the domain. By dopping a cetain numbe of coefficients in the domain, the enegy loss is calculated. Whethe o not sending the block to domain is based on the enegy loss. he simulation shows that the poposed Hybid - system povides good pefomance fo both inta fame coding and inte fame coding. Index ems adaptive vecto quantization,, inte fame, inta fame,, video coding.. INRODUCION ansfom coding is one of the coe algoithms in video coding. By applying diffeent linea tansfoms, the oiginal data ae decomposed into a set of coefficients in the coesponding tansfom domain. In the existing video coding standads such as H.64/AVC, H.63, MPEG, etc., discete cosine tansfom () plays the ole of tansfom coding due to its high enegy compaction and efficient computation complexity []. Singula value decomposition () is a tansfom that povides optimal enegy compaction fo any data [][4]. Howeve, the high computational complexity associated with computing the eigenvectos and eigenvalues has limited its application. Besides, only modest compession can be achieved by using because when eigenvalues ae tansmitted, the coesponding eigenvectos must also be tansmitted [9]. Dapena and Ahalt poposed a hybid algoithm fo image compession [6]. his algoithm can educe a lot computational Manuscipt eceived May 3, 6. ong, Lin (contact peso, Electical Engineeing Depatment, Univesity of exas at Alington, U.S., (email: lin_tong@hotmail.com) Rao, K.R., Electical Engineeing Depatment, Univesity of exas at Alington, U.S., (email: ao@uta.edu) 7 cost that occus by using only and also impove the image quality when only using. Based on the chaacteistics of video data, we popose a hybid - (H) technique fo video coding. It will take advantages of fist, and use only fo the blocks that does not compact enegy well. Adaptive vecto quantization (AV) technique is used to code the eigenvectos so that highe compession of can be achieved [7]. his pape is oganized as follows: Section intoduces coding techniques. Section 3 biefly eviews the most popula video coding technique and explains the motivation of the poposed H technique based on the chaacteistics of data extacted fom video compession. In Section 4, afte a shot eview of the existing H algoithms in image coding, we popose ou new video coding system. Section 5 shows the simulation esults. Conclusions ae pesented in Section 6.. CODING povides optimal enegy compaction fo any input data [8]. By taking only a few lagest eigenvalues and coesponding eigenvectos will give good epesentation of the input data. Matix A ( m has a singula value decomposition, which can be epesented by A = UΛ V () whee U is a m m othogonal matix, V is a n n othogonal matix, Λ is a m n matix whose off-diagonal enties ae all zeos. he diagonal elements of Λ must satisfy λ λ... λ λ+... λn, whee is the ank of A. he columns of V can be calculated by solving ( A λ ( I) v( = n =,..., () whee A = A A. he columns of U ae: u ( = Av( n =,..., (3) λ ( If q eigenvalues and coesponding eigenvectos ae chosen to epesent the oiginal

block, the econstucted block is: q Aˆ = λ ( u( v ( q (4) n= he squae eo is equal to the sum of the discaded eigenvalues: A( m, Aˆ( m, = λ ( (5) m= n= n= q+ whee A( m, is the oiginal pixel of the subblock and Aˆ ( m, is the econstucted pixel of the subblock. hus, the enegy of the econstucted subblock is the enegy in the q etained eigenvalues: Enegy = q n= λ ( (6) In conclusion, decomposes the oiginal data into two othogonal matices U and V, epesenting the eigenvectos, and a diagonal matix Λ epesenting the eigenvalues. Eigenvectos ae usually encoded by vecto quantization techniques, while eigenvalues ae encoded by scala quantization techniques. 3.INRODUCION OF VIDEO CODING AND CHARACERISICS OF VIDEO DAA Video data ae highly coelated data. In the most popula video coding techniques [3][3], the idea to compess video is to educe both spatial coelation and tempoal coelation of the oiginal data. In the existing video coding standads (like H.63, H.64, MPEG, MPEG4, etc), -dimentional has been adopted to educe spatial edundancy and inte-fame pediction is used to educe tempoal edundancy. When tempoal pediction is applied, we call the coding mode INER, as the compession techniques ae applied to pedict eos between adjacent fames; while if no tempoal pediction is applied, the coding mode is called INRA, as the edundancy eduction is applied to a single fame [3]. Fig. shows the block diagam of a typical video coding system. he standad deviation of a block descibes the pixel activity fo an INRA block o movement level fo an INER block. High standad deviation indicates many abupt changes in intensity in an INRA block o heavy motion occuence in an INER block, while low standad deviation means nealy unifom intensity in an INRA block o slight motion occuence in an INER block []. Applying and to blocks with diffeent standad deviations will give us diffeent esults. has the simila pefomance as when the standad deviation of the block is low, but when the standad deviation is high, pefoms much bette than [5]. 5 : 3 / 64 highest feq.coeff. :4 / 8 lagest E.val 45 4 35 3 5 5 5 5 3 4 5 6 7 8 : 6 / 64 highest feq.coeff. : / 8 lagest E.val 5 Fig.. Block diagam of a video coding system., Encode, Decode. 5 3 4 5 6 7 8 8

4 35 : 8 / 64 highest feq.coeff. : / 8 lagest E.val INER diffeential data can benefit fom the H technique. 3.4 Histogam of standad deivation 5. 5. Inta-Data Inte D ata 5 Pobability.8.6 3 4 5 6 7 8 (c) Fig.. Motivation analysis.,,(c), pefomance compaison between and by taking the same atio of coefficients. A seies of tests have been done to illustate this popety. We take 45 fames fom thee diffeent video sequences, Containe, Foeman, and News [6]. hey ae CIF sequences, with fame size 76x44, and 8 bpp. Each fame has been divided into blocks. 78 blocks ae used in this testing, and the standad deivation of the testing blocks is concentated in the ange [,8]. We applied and on these blocks. coefficients ae aanged in zigzag scan ode, while eigenvalues ae aanged in descending ode. In the thee expeiments, we dopped 5%, 5% and.5% of the lowest fequency coefficients in o smallest eigenvalues in domain espectively. hee is no quantization and coding involved. Compaing the econstucted block with the oiginal block, we get the fo each block. Fig. shows the esults in tems of mean squae eo vesus standad deviation of the block, when diffeent numbe of coefficients o eigenvalues /eigenvectos etained in the econstuction. It can be easily seen that fo blocks with standad deivation less than 6, both tansfoms can achieve small ; while fo the blocks with standad deivation geate than 6, the caused by both tansfoms become high. Compaatively speaking, pefoms bette than fo high standad deivation blocks. heefoe, it can be concluded that can be a compensating technique when cannot povide good compession quality..4. 4 45 5 55 6 65 7 75 Fig.3. Distibution of INRA data and INER diffeential data at diffeent standad deivations. 4. H VIDEO CODING SYSEM hee ae seveal appoaches to combine and in image compession [6][]. In these appoaches, standad deivation is used to choose between and. Wongsawat extended the H [6] into colo images and modified eigenvecto quantization method by applying adaptive multistage vecto quantization. o design the H system fo video compession, the appoaches need to match the chaacteistics of video data well. he oveall block diagam of the poposed H video coding system is shown in Fig 4. At the encode side, both the INRA data and the INER diffeential data go though the H code, eithe o coefficients will be tansmitted. A H decode is designed fo the pupose of econstuction. And the diffeence data between an inte fame and the pevious econstucted fame is pocessed by the H code. he motion data and heade infomation ae tansmitted diectly. he decode side is just the invese pocess of the encode side. H Encode o Coefficients H Decode o achieve high compession, motion compensation technique with the seach window 5x5 is applied. Geneally speaking, INER diffeential data will have elatively small standad deviation. While fo the blocks with heavy motion, the standad deviation can be vey high. Fig.3 shows the histogam of data in INRA mode and diffeential data in tempoal pediction INER mode. By obseving this distibution, it can be seen that both of them have high standad deviation blocks. heefoe both INRA data and 9 Motion Compensated Pedicto Motion Fig.4. H encode oveview Data he details of the poposed H encode and decode ae illustated in Fig.5.

8 x 8 Block Y Cb 6x6 C D No Subtact mean Calc. fo block < h Adaptive V S Yes Coeff. Evect. Eval. Mul t i pl exe Coded Bitsteam invese adaptive vecto quantization. U V yields the econstucted block. If the flag indicating is used, the quantize coefficients ae etieved. 3 3 9 8 5. SIMULAION RESULS Pefomance of H on inte fame video coding Coeff. D I 8 x 8 Block PSNR (db) 7 6 5 Coded Bitsteam De- Mul t i pl exe Evect. Eval. Adaptive V S Add Mean Rec. Stage 6x6 Y Cb C 4 3.5.5.5 35 Bit ate (bpp) H Pefomance of H on inta fame video coding Fig.5. Block diagams of H encode and decode. he system will nominally divide a video fame into 6x6 maco blocks. In each maco block, the data is oganized as fou luminance blocks and two chominance blocks. 4:: sampling patten is used. he decision of whethe o is chosen fo a paticula block is made in the domain. We compae the coefficients afte quantization and enegy dopping with the oiginal coefficients. If the is smalle than ou theshold, it means applying gives good esult at the given bit ate. In this case, the quantized coefficients ae tansmitted. he fist quantized coefficient is coded using 8 bits, and the emaining quantized coefficients ae coded using 6 bits, each. If the is geate than the theshold, the oiginal block will go though the banch. In the banch, the mean of the block is subtacted and unifomly quantized using 8 bits. We calculate the eigenvalues and select a cetain pecentage of enegy that is the same as the enegy etained in the domain. hen only a educed numbe of eigenvectos ae calculated. his deceases the computation cost. Scala quantization is applied to the eigenvalues and adaptive vecto quantization is applied to eigenvectos to achieve moe compession []. At the decode side, fo each block, the fist bit indicates which tansfom is used. If is used, the eigenvalues will be ecoveed by invese S. And matices U and V ae ecoveed by the 3 PSNR (db) 3 5.5.5.5 3 3.5 Bit ate (bpp) H Fig.6. Pefomance compaison of H and pue fo video sequence Caphone. fo INER coding fo INRA coding wo sets of simulations have been pefomed. One is fo INRA fame coding, the othe one is fo INER fame coding. he H video coding system is compaed with a pue video coding system, accoding to the atedistotion pefomance. he fomula to calculate distotion is: 55 PSNR = log db (7) N M = [ x( xˆ( ] (8) NM j= i= whee x ( is the oiginal value, and x ˆ( is the econstucted value. Figue 6 shows the esults fo INRA fame coding and INER fame coding espectively fo Caphone sequence. he bit ate in pat is contolled by adjusting the pecentage of enegy that has been dopped. 8 bits ae used to epesent the fist coefficient in each block, while 6 bits ae used fo the est of the

tansmitted coefficients. he bit ate in the H system is contolled by applying fo a diffeent numbe of blocks in a fame. Using equies high bit ate but povides good quality, while using equies lowe bit ate but povides degaded quality. he computational cost vaies at diffeent bit ate since diffeent pecentage of blocks is pocessed by. At vey low bit ate, the computational cost is as low as a pue system; while highe computational cost is needed when bit ate becomes highe as moe decomposition ae applied. It can be seen that H system has bette ate distotion pefomance than the pue system fo both inta fame coding and inte fame coding. 6. CONCLUSIONS his pape poposes a new H video coding system. If can compact the enegy of the block well, will be used; othewise, is applied. Adaptive V has been used to impove the compession in the banch. Simulation esults show that the poposed technique povides bette pefomance than a pue system fo both INRA fame coding and INER fame coding. Howeve, thee is limitation using fo data compession because when eigenvalues ae tansmitted, the coesponding eigenvectos must also be tansmitted. Although the poposed technique can outpefom a pue video coding system, it has limitation to outpefom a moe complicated based video coding system. Howeve, all the widely used video coding standads H.63, MPEG, H.64/AVC [3] integate many efficient compession tools togethe with tansfom coding and entopy coding, the poposed H technique does not seem to have the potential to outpefom those state of at video coding standads. REFERENCES [] Jain, A.K., Fundamentals of Digital Image Pocessing, Englewood Cliffs, NJ: Pentice Hall, 989. [] Ochoa, H., Rao, K.R., A low bit-ate hybid DW- image coding system (HDW) fo monochomatic images, IEEE, Symp. On Cicuits and Systems, ISCAS, 3, pp.47-475. [3] Rao, K.R., Wang, J.J., echniques and Standads fo Image Video and Audio Coding, Uppe Saddle Rive, NJ: Pentice Hall, 996. [4] Rao, K.R., Yip P.C., he ansfom and Data Compession Handbook, Boca Raton: CRC Pess,. [5] http://www.acticom.info/494.html. [6] Dapena, A., Ahalt, S., A hybid - image-coding algoithm, IEEE, ans. on CSV,, pp.4-. [7] K. Sayood, Intoduction to Data Compession. 3 d Edition, Mogan Kaufmann, 6. [8] Depenttee, Ed. F., and Signal Pocessing: Algoithms, Applications and Achitectues, New Yok: Noth Holland, 988. [9] Aase, S.O., Husoy, J.H. and Waldema, P., A citique of -based image coding systems, ISCAS 99, pp.3-6. [] Goldbeg, M., Bouche, P.R., and Shlien, S., Image Compession Using Adaptive Vecto uantization, IEEE ans. on Communications, 986, pp.8-87. [] Richadson, I.E.G., Video Codec Design, Chicheste, West Sussex: Wiley,. [] Wongsawat, Y, Ochoa,H, Rao, K.R., Oaintaa, S., A modified hybid - image-coding system fo colo image, ISCI, 4, pp.766-769. [3] Kwon, S., amhanka, A., Rao, K.R., Oveview of H.64/MPEG-4 Pat, J. Visual Communication and Image Repesentation, vol. 7, pp.83-55, Apil 6 ong, Lin was bon in Beijing, China in 975. She eceived the B.S. degee fom Univesity of Science and echnology of China, in Automation Depatment, in 999. Fom 999 to, she has been with the Legend Compute Company, Beijing, as a technical suppot enginee. She eceived the M.S. fom Alfed Univesity, NY, in Electical Engineeing in. Since, she has been woking towads the Ph. D. at the Univesity of exas at Alington. He doctoal eseach is on ate contol and H video coding. She woked as an inten in Philips Reseach Cente in 3 fo five months. Rao, K.R. eceived the Ph. D. degee in electical engineeing fom he Univesity of New Mexico, Albuqueque in 966. Since 966, he has been with the Univesity of exas at Alington whee he is cuently a pofesso of electical engineeing. He, along with two othe eseaches, intoduced the Discete Cosine ansfom in 975 which has since become vey popula in digital signal pocessing. He is the co-autho of the books Othogonal ansfoms fo Digital Signal Pocessing (Spinge-Velag, 975), Also ecoded fo the blind in Baille by the Royal National Institute fo the blind. Fast ansfoms: Analyses and Applications (Academic Pess, 98), Discete Cosine ansfom-algoithms, Advantages, Applications (Academic Pess, 99). He has edited a benchmak volume, Discete ansfoms and hei Applications (Van Nostand Reinhold, 985). He has coedited a benchmak volume, eleconfeencing (Van Nostand Reinhold, 985). He is co-autho of the books, echniques and standads fo Image/Video/Audio Coding (Pentice Hall) 996 Packet video communications ove AM netwoks (Pentice Hall) and Multimedia communication systems (Pentice Hall). He has coedited a handbook he tansfom and data compession handbook, ( CRC Pess, ). Digital video image quality and peceptual coding, (with H.R. Wu),aylo and Fancis (6). Intoduction to multimedia communications: applications, middlewae, netwoking, (with Z.S. Bojkovic and D.A. Milovanovic), Wiley, (6). Some of his books have been tanslated into Japanese, Chinese, Koean and Russian and also published as Asian (papeback) editions. He has been an extenal examine fo gaduate students fom Univesities in Austalia, Canada, Hong Kong, India, Singapoe, hailand and aiwan. He was a visiting pofesso in seveal Univesities -3 weeks to 7 and/ months- (Austalia, Japan, Koea, Singapoe and hailand). He has conducted wokshops/tutoials on video/audio coding/standads woldwide. He has supevised seveal students at the Mastes (59) and Doctoal (9) levels. He has published extensively in efeeed jounals and has been a consultant to industy, eseach institutes, law fims and academia. He is a Fellow of the IEEE. 3