The JPEG is dead! Long live the JPEG!
summary
Compares image formats with data: a 12,000 sample shows webp compresses ~33% smaller than older jpeg encoders; automated quality tuning using a structural similarity metric yields a surprising common setting near quality 59 and often only 0.13% dissimilarity to originals. Describes tools and techniques: MozJPEG/Modz jpeg, saliency maps and per-tile adaptation, PNG Quant 2, svgo, low‑quality placeholders (elip/esip), and JPEG XT’s backward‑compatible HDR/alpha proposal.
AI-written summary · about
chapters (13)
- 0:00Start
- 5:09Who won the race?
- 6:54Bandwidth is cheap
- 11:07JPEG: built based on human visual system
- 15:22Cones are for colors: long, medium & short wavelength cone types
- 16:37Huffman table optimization
- 18:28Optimized Huffman tables
- 19:19Custom quantization matrices
- 19:30Trellis quantization
- 19:43Optimized progressive encoding
- 27:57MSSS Maximum Symmetric Surround Saliency Algorithm
- 33:16JPEG XT
- 43:40TAKEAWAY
related talks
47mImage Optimization
12mExtreme Image Optimisation: WebP & JPEG XR
03mDelivering fast and beautiful images and video
01h 11mMoving Pictures
26mFast and Beautiful Modern Image Delivery Techniques
48mFast and Beautiful: Modern Image Delivery Techniques
37mImage Formats, Performance and Cognitive Load
36mFast and Beautiful: Modern Image Delivery Techniques