CS4624 MPEG

A great deal of information about MPEG can be learned by studing Frequently Asked Questions.

You also may wish to refer to a PDF version of the standard.

Overview of the ISO/IEC standard MPEG (Motion Pictures Expert Group):

• Approval: video standard specified in 1991, audio and then systems later
  
• Adoption:
  
  
  • CD-I add-on
    
  • 3DO add-on
    
  • Sigma Designs Reel Magic CD-ROM Upgrade Kit
  
  
• MPEG II:
  
  
  • Higher resolution (to HDTV) and hence bitrate 704 x 480 at 30 fps
    
  • Direct satellite broadcasts, cable TV
  
  
• MPEG details:
  
  
  • Interframe coding has (I) reference frames, and predicted frames going forward (P) as well as bi-directional (B)
    
  • Decoding includes handling of motion vectors for inter-frame coding and inverse DCT and Huffman for intra-frame coding
    
  • Interframe decoding also involves interpolation
    
  • Net effect: single-speed CD-ROM, 320x240, 30 fps
  
  

• Filtering: Simple approach is decimation; Better approach is averaging, using multiple taps (pixels/lines)
  
• Color-space conversion (MPEG compression 1.5:1): RGB to/from YUV
  
• Digitizing (2:1)
  
• Scaling or subsampling (4:1):
  
  
  • Pick resolution and sample rate / sampling; YUV studio quality is 4:2:2 while MPEG and DVI use 4:1:1
    
  • QuickTime originally to 160x120, MPEG to 320x240
    
  • Also, pixel depth reduction in number of colors (CLUT)
    
  • Playback to double or quadruple: smart scaling interpolation
  
  
• Transforms: DCT, Wavelet
  
• Quantization (3:1):
  
  
  • Different accuracy of representation for different parts
    
  • Less detail for high frequency due to Human Visual System
  
  
• Entropy coding (1.5:1):
  
  
  • Huffman or arithmetic
    
  • Runs of zeroes
  
  
• Interframe (5:1):
  
  
  • Predictive: Store difference values instead of actuals
    
  • Motion estimation and compensation: Find motion vector indicating pixel offset of 16x16 macroblock
  
  
• Picture interpolation:
  
  
  • Average prior and following to predict current
    
  • Selectively use prior or following to predict each region
    
  • Save difference from predicted value