Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Lossy PNG for true-color images (Velocity Conf ...
Search
Kornel Lesiński
November 18, 2014
Design
2
900
Lossy PNG for true-color images (Velocity Conf EU 2014)
Kornel Lesiński
November 18, 2014
Tweet
Share
More Decks by Kornel Lesiński
See All by Kornel Lesiński
SudWeb
kornel
0
210
Modern PNG compressors
kornel
13
1.2k
Handling binary data in JS
kornel
1
400
Mobblestouch — W3C Touch Events and gestures
kornel
2
730
Server-Sent Events
kornel
5
980
"HTML5" Transport Protocol
kornel
3
400
Front-end Performance
kornel
5
500
Other Decks in Design
See All in Design
「余白」と「欲望」を味方につける ——AI時代のデザインエンジニアリングと「越境」の作法 #KNOTS2026
koyaman
1
1.2k
Correspondence:共に生成していく過程
akiramotomura
0
180
デザイナーが主導権を握る、AI協業の本音と実践
satosio
7
2.5k
Emmy's Artwork
mcksmith
0
190
TUNAG BOOK 2024
stmn
PRO
0
1.3k
公開スライド)熊本市様-電子申請中級編
garyuten
0
730
mount_company_profile
mount_inc
0
4.7k
生成AIの不確実性を価値に変える、「ビズリーチ」の体験設計 / KNOTS2026
visional_engineering_and_design
4
570
組織の右腕として共創する ー デザインと経営の二つの視点から見えた、新しい支援のかたち/ Designship2025_Nishimura
root_recruit
0
280
Ralph Penel Portfolio
ralphpenel
PRO
0
250
AI業務アプリケーションの体験デザイン
kazuhirokimura
0
250
逆向きUIの世界〜AndroidアプリのRTL言語対応〜
akatsuki174
1
780
Featured
See All Featured
Taking LLMs out of the black box: A practical guide to human-in-the-loop distillation
inesmontani
PRO
3
2k
Music & Morning Musume
bryan
47
7.1k
The AI Revolution Will Not Be Monopolized: How open-source beats economies of scale, even for LLMs
inesmontani
PRO
3
2.9k
The MySQL Ecosystem @ GitHub 2015
samlambert
251
13k
Marketing Yourself as an Engineer | Alaka | Gurzu
gurzu
0
120
職位にかかわらず全員がリーダーシップを発揮するチーム作り / Building a team where everyone can demonstrate leadership regardless of position
madoxten
55
49k
Organizational Design Perspectives: An Ontology of Organizational Design Elements
kimpetersen
PRO
1
71
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
55
3.2k
Leveraging Curiosity to Care for An Aging Population
cassininazir
1
150
The Anti-SEO Checklist Checklist. Pubcon Cyber Week
ryanjones
0
53
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
66
Optimising Largest Contentful Paint
csswizardry
37
3.6k
Transcript
Lossy PNG Kornel Lesiński 2014
PNG Lossless JPEG Lossy
PNG PNG JPEG Lossless Lossy PNG Opaque Alpha Channel WebP
JPEG 2000 JPEG XR JPEG XT if you need lossy and alpha, it becomes complicated. You may need to serve 4 different files for every image!
flickr.com/photos/ama711/8760361969 0% 100% Lossy PNG is a hack, but it
works like a truly lossy format
Pixels Math gzip Standard PNG
None
gzip finds and removes repeated sequences of identical pixels
00000000 00000001 00000010 00000011 00000100 00000101 00000110 00000111 00001000 00001001
00001010 0 10 110 1110 11110 111110 1111110 11111110 111111110 1111111110 11111111110 → → → → → → → → → → → Instead of storing 8 bits per byte, gzip uses variable-length sequences. Most popular bytes are given shortest sequences.
Some images don’t have repeated pixels and few very popular
colors
0 1 2 3 4 5 6 7 8 9
0 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 = = Instead of storing uncompressible absolute values PNG can store difference between neighboring pixels
5 6 4 6 5 5 6 4 5 4
5 + 1 -2 +2 - 1 0 + 1 -2 + 1 - 1 = = 5 0 0 0 0 0 0 0 0 0 0 Noisy images won’t compress well… unless you fake the data and remove the noise
Pixels Math gzip Lies Lossy PNG trick is about replacing
math with lies. Only the encoder changes, so it’s 100% compatible.
? When PNG “guesses” what the next pixel will be,
it can make the guess based on average of 2 neighbors, which in lossy PNG causes blur
flickr.com/photos/tundra-ice/5124841152 Original: 2MB 0.6MB 0.3MB
Original: 2MB 0.6MB 0.3MB
Original: 221KB Lossy: 67KB ed_g2s
182KB flickr.com/photos/_tar0_/6223343149
27KB flickr.com/photos/_tar0_/6223343149 The algorithm can decide quality of every pixel,
so it tends to preserve sharp edges, especially text
28KB flickr.com/photos/_tar0_/6223343149 Top of the file has 0% quality, bottom
80% quality
• Works everywhere—fully backwards-compatible • No color limit—supports 24 and
32-bit PNG • Preserves edges—per-pixel quality control • Very fast—small tweak in the encoder • Streamable—only needs 2 lines of the image
pngquant -Q 75-100 --skip-if-larger \ -o out.png in.png || posterize
-b -Q 75 in.jpg out.jpg pngquant is often more efficient, but can’t guarantee quality. You can combine the two for best results.
pngmini.com — ImageAlpha GUI github.com/pornel/mediancut-posterizer CLI logicnet.dk/Analyzer — Image Analyzer
GUI @pornelski pngmini.com/lossypng.html