Compression Techn的iques of Multimedia_3.pptVIP

Compression Techniques of Multimedia Dr. Xiqun Lu College of Computer Science Zhejiang University Outline Quantization Distortion (Error) Image Compression Entropy Huffman Coding Scalar Vector Quantization Transform compression JPEG Video Compression Motion estimation MPEG Outline Quantization Distortion (Error) Image Compression Entropy Scalar Vector Quantization Transform compression JPEG Video Compression Motion estimation MPEG Quantization Distortion Because of the quantization, there is some distortion between the actual source sample xk and the corresponding quantized values Q{xk}: d(xk, Q{xk}) = |xk – Q{xk}| The distortion between a sequence of n samples Xn and the corresponding n quantized values Q{Xn} is the average over the n source output samples d(Xn, Q{Xn}) = 1/n ?d(xk, Q{xk}) Outline Quantization Distortion (Error) Image Compression Entropy Huffman Coding Scalar Vector Quantization Transform compression JPEG Video Compression Motion estimation MPEG Entropy In information theory, it is helpful to view data samples from a probabilistic perspective. We define information as the uncertainty with each data sample. Suppose we have a set of discrete-time, discrete-valued random variables {s1, s2, …, sN}, and each data sample with a probability P(si) (i = 1, 2, … N) Self-information of each data sample is defined as The average information rate of the source is given by the entropy (measured by bits) Entropy Coding Entropy codes can be used to code at rates very close to the source entropy. A straightforward way to this problem is to let the variable -codeword length of each data sample si be Prefix Condition No codeword in the set may be the prefix of another codeword to insure uniquely and instantaneously decode-able Huffman Coding Optimal Variable Length Coding 1) To order the symbols in decreasing order of probability. 2) To merge the symbols with the lowest probabilities, and reordering the new branches in decreasing order of probability, repeat th