@@ -54,12 +54,14 @@ Name | Option | Description
54
54
----------- | ------------- | -----------
55
55
MPE | ` -m mpe ` | Mean pixel error (as used by [ imgmin] ( https://github.com/rflynn/imgmin ) )
56
56
PSNR | ` -m psnr ` | [ Peak signal-to-noise ratio] ( https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio )
57
+ MSE | ` -m mse ` | [ Mean squared error] ( https://en.wikipedia.org/wiki/Mean_squared_error )
58
+ MSEF | ` -m msef ` | ` sqrt(MSE/Variance) `
59
+ Correlation | ` -m cor ` | [ Correlation] ( https://en.wikipedia.org/wiki/Correlation )
57
60
SSIM | ` -m ssim ` | [ Structural similarity] ( http://en.wikipedia.org/wiki/Structural_similarity )
58
61
MS-SSIM* | ` -m ms-ssim ` | Multi-scale structural similarity (slow!) ([ 2008 paper] ( http://foulard.ece.cornell.edu/publications/dmr_hvei2008_paper.pdf ) )
59
62
VIFP1 | ` -m vifp1 ` | [ The visual information fidelity (VIF)] ( http://live.ece.utexas.edu/publications.php ) 1 layer.
60
63
SmallFry | ` -m smallfry ` | Linear-weighted BBCQ-like ([ original project] ( https://github.com/dwbuiten/smallfry ) , [ 2011 BBCQ paper] ( http://spie.org/Publications/Proceedings/Paper/10.1117/12.872231 ) -> [ LibSmallFry] ( https://github.com/ImageProcessing-ElectronicPublications/libsmallfry ) )
61
64
SharpenBad | ` -m shbad ` | Sharpen discrepancies ([ LibSmallFry] ( https://github.com/ImageProcessing-ElectronicPublications/libsmallfry ) )
62
- Correlation | ` -m cor ` | [ Correlation] ( https://en.wikipedia.org/wiki/Correlation )
63
65
NHW | ` -m nhw ` | NHW convolutional metric ([ original project] ( https://github.com/rcanut/NHW_Neatness_Metrics ) -> [ LibSmallFry] ( https://github.com/ImageProcessing-ElectronicPublications/libsmallfry ) )
64
66
1 pair | ` -m ssimfry ` | ` (ssim + smallfry) / 2 `
65
67
2 pair | ` -m ssimshb ` | ` (ssim + shbad) / 2 `
@@ -73,15 +75,16 @@ SUMMARY | `-m sum` | `(ssim + vipf1 + smallfry + shbad + nhw) / 5` **DE
73
75
```
74
76
Trends:
75
77
```
76
- UM = 2.99 * sqrt(sqrt(1.0 / MPE)) - 1.70
77
- UM = 1.00 * sqrt(PNSR) - 5.32
78
- UM = 2.07 * cor_sigma(cor_sigma(cor_sigma(SSIM))) - 0.26
78
+ UM = 2.42 * sqrt(sqrt(1.0 / MPE)) - 1.38
79
+ UM = 0.87 * sqrt(PNSR) - 4.70
80
+ UM = 1.02 * sqrt(sqrt(1.0 / MSEF)) - 1.49
81
+ UM = 2.87 * cor_sigma(cor_sigma(COR)) - 1.42
82
+ UM = 1.73 * cor_sigma(cor_sigma(cor_sigma(SSIM))) - 0.11
79
83
UM = 1.59 * cor_sigma(cor_sigma(MS_SSIM)) + 0.01
80
- UM = 3.69 * cor_sigma(VIFP1) - 2.74
81
- UM = 0.0684 * SMALLFRY - 6.29
82
- UM = 1.17 * SHARPENBAD - 0.12
83
- UM = 3.03 * cor_sigma(cor_sigma(COR)) - 1.52
84
- UM = 0.40 * sqrt(sqrt(1.0 / NHW)) - 0.48
84
+ UM = 1.12 * cor_sigma(cor_sigma(VIFP1)) - 0.03
85
+ UM = 0.0658 * SMALLFRY - 6.07
86
+ UM = 1.00 * SHARPENBAD + 0.05
87
+ UM = 0.36 * sqrt(sqrt(1.0 / NHW)) - 0.41
85
88
86
89
cor_sigma(M) = 1.0 - sqrt(1.0 - M * M)
87
90
```
@@ -92,7 +95,7 @@ The JPEG format allows for subsampling of the color channels to save space. For
92
95
93
96
#### Example Commands
94
97
95
- ``` bash
98
+ ``` shell
96
99
# Default settings
97
100
jpeg-recompress image.jpg compressed.jpg
98
101
0 commit comments