SNAKE OIL
IN IMAGE
COMPRESSION
Kornel Lesiński
London Web Performance 2016
Slide 2
Slide 2 text
=
WITHOUT LOSS
=
QUALITY 100%
=
IRREVERSIBLY
LOSES
QUALITY
LOSSY
LOSSLESS
Slide 3
Slide 3 text
QUALITY DEPENDS
ON
VIEWING CONDITIONS
(CONTRAST, PIXEL SIZE)
Slide 4
Slide 4 text
The basic idea behind
lossy compression is that
eyes are imperfect and
humans don't see every
detail, so dototε tnot oro
too εnott to bo εoon, εon
bo roplocod nitn o roμpn
opproεnoton.
Slide 5
Slide 5 text
FLIF
GIF
Slide 6
Slide 6 text
WORST LOSSY FORMAT
IS STILL SMALLER THAN
THE BEST LOSSLESS
Slide 7
Slide 7 text
FLIF
GIF
230KB
21KB
Slide 8
Slide 8 text
GIF
21KB
JPEG
20KB 30KB?
Slide 9
Slide 9 text
QUALITY
FILE SIZE
Slide 10
Slide 10 text
QUALITY
FILE SIZE
Red
Blue
Slide 11
Slide 11 text
QUALITY
FILE SIZE
Red
Blue
Slide 12
Slide 12 text
COMPARE AT
EXACTLY
THE SAME FILE SIZE
Slide 13
Slide 13 text
QUALITY
FILE SIZE
Red
Blue
Slide 14
Slide 14 text
QUALITY
FILE SIZE
Red
Blue
Slide 15
Slide 15 text
BENCHMARK
ONLY
USEFUL QUALITY
RANGE
Slide 16
Slide 16 text
QUALITY
FILE SIZE
Red
Blue
Slide 17
Slide 17 text
QUALITY
Slide 18
Slide 18 text
0 100
Slide 19
Slide 19 text
"QUALITY" SETTING
IS MADE UP
Slide 20
Slide 20 text
No content
Slide 21
Slide 21 text
0 100
Slide 22
Slide 22 text
Difference 1.1
Difference 3.5
Slide 23
Slide 23 text
PSNR (dB) / MSE
Peak Signal To Noise Ratio / Mean Square Error
➤ Simple per-pixel difference
➤ Inaccurate, easy to game
➤ Loves blur, hates sharpness
Slide 24
Slide 24 text
SSIM
Structural Similarity
➤ Quite decent
➤ Aware of noise and edges
➤ Score is tool-dependent
➤ MS-SSIM, HVS-PSNR, Butteraugli