MPEG + Compression of Moving Pictures for Digital Cinema Using the MPEG-2 Toolkit. A Digital Cinema Accelerator

Transcription

1 142nd SMPTE Technical Conference, October, 2000 MPEG + Compression of Moving Pictures for Digital Cinema Using the MPEG-2 Toolkit A Digital Cinema Accelerator Michael W. Bruns James T. Whittlesey 0

2 The Challenge Leverage existing standards to accelerate practical application of compression for Digital Cinema A viable choice is MPEG + -- new usage of MPEG-2 toolkit that meets today s requirements for digital cinema Plus -factor includes enhancements to address image quality MPEG factor addresses interoperability, flexibility, reliability MPEG-2 is fundamentally a rich compression tool set that can applied beyond the limits of its common usage 1

3 Why Compress for Digital Cinema? Year 2000 digital cinema presentation: 120 minutes 60 seconds per minute 24 frames per second 1080 lines per frame 1920 pixels per line X 3 samples per pixel trillion samples Visually transparent compression is possible 2

4 Image Compression for Digital Cinema Image includes tangible and intangible elements colors, film grain are part of the artistic presentation Compression must preserve all elements without distortion 3

5 Digital Cinema Compression Options? Advanced MPEG-based Algorithms Lossless Wavelets Proprietary Algorithms JPEG-2000 Criteria for selection picture quality real-time encoding and decoding standards basis 4

6 How MPEG-2 Standard is Used Today Compression techniques, bitstream syntax = Toolkit Discrete cosine transform quantization Modified Huffman variable length coding motion compensation Profiles and Levels, Constraints = Usage Rules Main Profile => 4:2:0, YCbCr color space high level - for HDTV main level - for internet and DVD, satellite distribution Applications = Additional Usage Rules ATSC Mb/s constant bit rate Internet: severe constant bit rate constraint 5

7 Practical Application of MPEG-2 for Digital Cinema Same toolkit Define new usage rules for Digital Cinema Constant Quality Rate Control consistent, predictable picture quality 4:4:4, Equal Color Bandwidth, RGB Color Space expanded palette of colors Error Diffusion gives the effect of greater bit depth Allows use of standard silicon devices 6

8 Constant Quality Rate Control What is it? Constant Bit Rate Made for limited through-put distribution channels Encoder automatically changes its own loss characteristics to meet bit-rate allowance... Regardless of complexity or quality expectations Most commonly used Constant Quality Variable bit rate Quantization ( Loss ) characteristics strictly defined and unchanging Different bit-rate per scene depending on complexity Not commonly used The native rate control mechanism for MPEG-2 7

9 MPEG-2 Compression Tools Block DCT => Quantization => Variable Length Coding What is it? Why do it? Converts spatial image block to frequency domain Reduces number of bits required to represent image Human visual system is insensitive to some quantization depending on spatial frequency A form of entropy coding Common values i.e. small numbers translate to short codes 8

10 MPEG-2 Tools Example Image Block 8 bit samples Image Block luma as height DCT coefficients 12 bits F[u,v] 9

11 Quantization Divide each DCT coefficient by a constant factor and a frequency dependent factor QF[ u, v] = 16 F[ u, v] MQUANT W[ u, v] Define normalized quantization matrix w[ u, v] = MQUANT 16 W[ u, v] 1 => no loss 10

12 Quantization = DCT coefficients 12 bits F[u,v] Normalized quantization matrix w[u,v] Quantized DCT coefficients QF[u,v] 11

13 Decoder Function Original Inverse Quantization Reconstructed Image Block 12

14 Effect of w[u,v] on a sequence of pictures Case 1: Fix W[u,v], vary MQUANT (a) (b) (c) bits/pixel (a) (b) (c) lossless picture 13

15 Effect of w[u,v] on a seqence of pictures Case 2: Vary W[u,v], fix MQUANT (a) (b) bits/pixel (a) (b) picture 14

16 Color Space Advantages of Equal Color Bandwidth (4:4:4) RGB No chroma errors on (sharper than nyquist) edges Native colorspace of Telecine and Projector Fewer colorspace conversions May be recommended for Digital Cinema Distribution Master More colors for a given bit depth than Y Cb Cr Advantages of 4:2:2 or 4:2:0 Y Cb Cr More efficient use of uncompressed channel and compressed bits Which is better for digital cinema Depends on application and picture Film transfer versus rendered graphics Choice 15

17 Color Space (0, 0.5, 0.5) (1, 1, 1) G Y R C r (0, 0, 0) B (0, -0.5, -0.5) C b RGB cube within Y Cb Cr cube 75 % of Y Cb Cr volume does not correspond to a color 16

18 Error Diffusion Technique borrowed from printing Eliminates banding/contouring artifacts by giving the effect of greater bit depth Quantization errors are distributed about neighboring pixels Not random noise 17

19 Error Diffusion: Visual Example Shallow Ramp: 6 bit RGB Shallow Ramp: 6 bit RGB with error diffusion 18

20 Error Diffusion: Numerical Example 11 bit samples before error diffusion bit samples after error diffusion

21 New Usage Rules MPEG-2 is a rich set of compression tools that may be adapted for the quality expectations of digital cinema Constant Quality Rate control Equal Color Bandwidth RGB (4:4:4) colorspace Error Diffusion 20

22 Standards Basis Standards provide key success factors Unequivocal bitstream specification allows conformance checking Interoperability between manufacturers competitive market for capital equipment no single source Leverage devices and components across multiple markets lower component pricing through higher volumes Enable alternate program material live sports events, etc. 21

23 Conclusion MPEG+ compression enables digital cinema distribution today Picture quality Realtime Encoders and Decoders Standards Basis 22

24 With Permission 23