多媒体信号与处理小结(2).ppt

ECTE401 Multimedia Signal Processing,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,1,Revision Lecture,Lecture 1,Described multimedia and multimedia signalsTime and frequencyAnalog and digitalSampling QuantisationUniform,non-uniform,adaptive,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,2,Lecture 2,Define CompressionDescribe lossless and lossy compressionLossless CompressionFixed length binary codeHuffman code Entropy rate,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,3,Lecture 3,Speech signals:Voiced and unvoicedShort term correlationLinear predictionRemoves short term correlationModels the vocal tractG721 predictive speech coder32 kbps backward adaptive speech coderEvaluating speech qualityMOS tests,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,4,Lecture 4,Introduction to speech coding standardsReview of LP based speech compression modelsExamination of representing the excitation signalReview of current LP based speech coding standards,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,5,Lecture 5,The human auditory systemCD AudioMPEG Audio codingMPEG-1 Layer III(MP3)MPEG-2MPEG-4Dolby Digital AC3Multichannel audio codingMPEG Surround,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,6,Lecture 6,Image RepresentationDigital imagesColour PerceptionVideo representationBasics of Image and Video Coding,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,7,Lecture 7,Transforms and image codersWhy transform?Discrete Cosine TransformBaseline JPEG,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,8,Lecture 8,Discrete Wavelet TransformJPEG2000,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,9,Lecture 9,Motivation for video compressionDifferential encodingMotion EstimationMotion Compensation,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,10,Lecture 10,MPEG-1 and MPEG-2 Video Coding:Form frames into a GOPUse motion estimation and compensationUse JPEG based transformation and quantisationMPEG-2 Video Coding:Provides many more options and bit ratesAllows field based video codingMPEG-4 Video Coding:describes video using a sequence of video objectsUses a hierarchical descriptionEncodes shapes,textural and motion informationAllows scalable encoding and user interactionH.264 Video Coding,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,11,Typical Exam Questions:Speech and Audio Coding,1.is Analysis-by-Synthesis as used in speech coding and what advantage does it provide?See Lecture 4(e.g.slide 14 onwards).What is the purpose of Linear Prediction?Model the short term correlation of the speech signal.See Lecture 3.,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,12,Typical Exam Questions:Speech and Audio Coding,3.Using Huffman coding,find the codewords for the following table of symbols and probabilities.Calculate the entropy of the source and compare with the average codeword length resulting from Huffman coding.Entropy=2.2391Average codeword length=2.33 bits,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,13,Typical Exam Questions:Speech and Audio Coding,4.List and briefly describe the main steps involved in performing LPC analysisSee Lecture 4Find the mean squared error between the original speech and predicted speechMinimise the mean squared error by setting the derivative to 0Form a matrix representation based on the autocorrelation of the speech signalSolve for the LPC coefficients using Levinson-Durbin recursion,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,14,Typical Exam Questions:Speech and Audio Coding,5.List and briefly describe the main steps involved in performing LTP analysis.See Lecture 4Find a current segment of the residual signalCompare this segment with a previous segment located at a delay M multiplied by a gain value Find the minimum mean squared error between the delayed and current residual segment by setting the derivative to zero and solving for the gain valueRepeat for all possible values of M and choose M and that give the smallest mean squared error between the current and delayed segment,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,15,Typical Exam Questions:Speech and Audio Coding,6.What is the main objective when designing a speech coder?Ensure the perceptual quality is maximised7.Draw a block diagram of a CELP speech coder.Briefly describe each block.See Lecture 4,slide 288.Explain how the speech production process is modelled in a speech coder.See slide 6,Lecture 4,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,16,Typical Exam Questions:Speech and Audio Coding,9.What is the Equal Loudness Curve?See slide 5,lecture 5.The SPL required at each frequency so that a listener perceives the sound to be of equal loudness.10.What is masking?See slide 8,lecture 511.Draw a block diagram showing the main stages of a perceptual audio coder.See slide 14,lecture 5.,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,17,Typical Questions:Image and Video Coding,1.Draw and label the main blocks of a transform-based digital image coderSee slide 25 of Lecture 6 or for more detailed answer slide 10 of Lecture 72.Draw and label the main blocks of a transform-based digital video coder.What is the key difference between a video coder and an image coder?Similar answer to question 1,but remember to add the motion estimation and compensation block,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,18,Typical Questions:Image and Video Coding,3.What is the difference between the storage requirements for black and white images and colour images?Black and white images require more bits4.Compare how the DCT achieves scalable coding with how the DWT achieves scalable codingDCT:discard low energy coefficientsDWT:add more filtering stages,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,19,Typical Questions:Image and Video Coding,5.Assume a frame size of 256x256 pixels and a motion compensated video coder is using block sizes of 8x8 pixels with a maximum displacement of each block being 7 pixels in any direction.Assume also that the prediction error is quantised using 0.5 bpp.Compare the bit rates(in bpp)for one P-frame using this technique compared with JPEG compression at 2 bpp.First find the number of motion vectors=256x256/(8x8)=1024Next,find the number of bits/motion vector=4+4=8Next,find the number of bits per frame=8x1024+0.5x256x256=40960 bitsCompare with 2 bpp number of bits=2x256x256=131072 bits,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,20,Typical Questions:Image and Video Coding,6.I wish to quantise a sequence of numbers that are highly correlated.What is the advantage of differential quantisation of these numbers compared with direct quantisation of each number?Differential quantisation results in a lower bit rate,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,21,Typical Questions:Image and Video Coding,7.Draw a block diagram illustrating the main components of an MPEG-2 video coderSee lecture 10.Need to include transform,motion estimation and quantisation,quantisation and lossless coding.,Revision Lecture,Copyright 2008-2012,Christian Ritz,University of Wollongong,22,