jpeg 1991 gk wallace paper on jpeg standard
DESCRIPTION
A review of the 1991 Wallace's paper on JPEG standardTRANSCRIPT
The JPEG S*ll Picture Compression Standard
by Gregory K. Wallace, 1991
Julià Minguillón
Barcelona Compsci Club
Contents • Why a talk on JPEG? • Previous to JPEG • Technology at that Fme • The paper • Math foundaFons • The algorithms • Impact • Caveats • Related papers
Why a talk on JPEG?
• JPEG images are everywhere! – Facebook: 350000000 photos uploaded every day – Almost 1000000000000 photos taken every year!
• 1995 capstone project at UAB – ImplementaFon of progressive operaFon modes in IJG library v6b
• Perfect excuse for checking over Lena again!
Lena 512 x 512 x 24bpp
hZp://sipi.usc.edu/database/?volume=misc
Previous to JPEG
• … • TIFF 1986 • GIF 1987 • PBM 1991 • … • Limited compression raFos (lossless) • No standard for lossy image compression • JPEG commiZee created in 1986 (ISO in 1983)
Technology at that Fme
• 1991: Kodak DCS 100, 1.3MP, 200MB HD, only 987 sold (aimed at photojournalists, 20000$)
• 1994: Apple QuickTake 100, 0.3MP (VGA), 749$ • … • Computers in 1991: Intel 486DX, MS-‐DOS 5.0, Windows 3.0, Linux, VGA (640x480x256 -‐ yes, Amiga was beZer!), Adobe Photoshop 2.0, 1st color image scanner (HP)
The paper
Wallace, G.K. (1991). The JPEG sFll picture compression standard. CommunicaFons of the ACM – special issue on digital mulFmedia systems, 34(4), 30-‐44. • Summary of JTC1/SC2/WG10 • Already a classic (> 4500 cites) hZp://www-‐ee.eng.hawaii.edu/~treed/EE416/Project_4/jpeg-‐wallace.pdf
Math foundaFons
• Lossy compression ≈ remove what eye can’t see • Karhunen-‐Loève transform: any stochasFc process can be decomposed in an infinite linear combinaFon of orthogonal funcFons – Coefficients are decorrelated – Any truncaFon minimizes MSE
• Discrete cosine transform – Similar to DFT but real coefficients – Existence of fast / reversible / integer versions
The algorithms
• Color conversion from R-‐G-‐B to Y-‐Cb-‐Cr • Sampling and interleaving • DCT • QuanFzaFon (loss/compression happens here) • Reordering (zig-‐zag sequence) • Entropy coding: zeros, markers
• Symmetrical process for coding / encoding
Impact
• All devices taking pictures implement JPEG • JPEG File Interchange Format (.jpg, .jpeg) • IJG library used elsewhere • Hardware encoders / decoders available • Part also of M-‐JPEG (prior to MPEG-‐4) • Superseded by JPEG2000 (but…)
Caveats
• 8 x 8 blocks are too small • “Limited” to 12 bit samples • Bad quality for high compression raFons • ArithmeFc coding not used due to patents • HVS opFmizaFon barely used • In general, most devices/apps do not allow users to specify all JPEG parameters
• Awful results if wrongly used (text, diagrams)
Related papers Rao, K.R., Yip, P. (1990). Discrete Cosine Transform algorithms. Advantages, ApplicaFons. Academic Press, Inc. London. Huffman, D.A. (1962). A method for the construcFon of minimum redundancy codes. In Proceedings of the IRE, 40, 1098-‐1101. Pennebaker, W.B., Mitchell, J.L. (1993). JPEG SFll image data compression standard. Van Nostrand Reinhold, New York. Peterson, H.A., Ahumada, A.J., Watson, A.B. (1993). An Improved DetecFon Model for DCT Coefficient QuanFzaFon. In Human Vision, Visual Processing, and Digital Display IV, ed. B.E. Rogowitz and J. Allebach, Proc. 1913, 191-‐201. Minguillón, J., Pujol, J. (2001). JPEG standard uniform quanFzaFon error modeling with applicaFons to sequenFal and progressive operaFon modes. Journal of Electronic Imaging, 10(2), 475-‐485.