google / butteraugli

butteraugli estimates the psychovisual difference between two images
Apache License 2.0
1.96k stars 137 forks source link

Is there any suggest about the butteraugli compare result value? #22

Open shiyun20 opened 7 years ago

shiyun20 commented 7 years ago

I using butteraugli compare two image, the result value is "0.971608", according the result value, how can I know the two image match is perfect or ok or bad?

cvley commented 7 years ago

@shiyun10 , I have seen a post said that the value less than 1.1 is perfect. And in the guetzli project, 1.0 is used. Just for your information.

khavishbhundoo commented 7 years ago

@shiyun10 Less than 1.0 is best , 1.0-1.1 is acceptable , greater than 1.1 (not good).

AlexVonB commented 5 years ago

I found more information on that in the guetzli (jpeg compression optimization) repo: https://github.com/google/guetzli/blob/b473cf61275991e2a937fe0402d28538b342d2f8/guetzli/quality.cc#L26

// Butteraugli scores that correspond to JPEG quality levels, starting at
// kLowestQuality. They were computed by taking median BA scores of JPEGs
// generated using libjpeg-turbo at given quality from a set of PNGs.
// The scores above quality level 100 are just linearly decreased so that score
// for 110 is 90% of the score for 100.
const double kScoreForQuality[] = {
  2.810761,  // 70
  2.729300,
  2.689687,
  2.636811,
  2.547863,
  2.525400,
  2.473416,
  2.366133,
  2.338078,
  2.318654,
  2.201674,  // 80
  2.145517,
  2.087322,
  2.009328,
  1.945456,
  1.900112,
  1.805701,
  1.750194,
  1.644175,
  1.562165,
  1.473608,  // 90
  1.382021,
  1.294298,
  1.185402,
  1.066781,
  0.971769,  // 95
  0.852901,
  0.724544,
  0.611302,
  0.443185,
  0.211578,  // 100
  0.209462,
  0.207346,
  0.205230,
  0.203114,
  0.200999,  // 105
  0.198883,
  0.196767,
  0.194651,
  0.192535,
  0.190420,  // 110
  0.190420,
};