stb-image-metrics

This utility compares two graphics files based on metrics.

Includes metrics:

The value of MSE metrics is from 0.0 (minimum difference) to 1.0 (maximum difference).

The value of PSNR and SDSNR metrics is from 0.0 (maximum difference) to inf.

The value of SSIM and VIFP1 metrics is from 1.0 (minimum difference) to 0.0 (maximum difference).

The value of SHARPENBAD metrics is from 1.0 (minimum difference) to 0.0 (maximum difference).

The value of Correlation metrics is from 1.0 (minimum difference) to 0.0 (maximum difference).

The value of NHW metrics is from 0.0 (minimum difference) to 1.0 (maximum difference).

build

load submodules

submodules:

stb -> src/stb

$ git submodule init
$ git submodule update

install dependencies

build dependencies:

build-essential
cmake

$ sudo apt-get install build-essential cmake

compilation

$ mkdir build
$ cd build
$ cmake ..
$ make

use

The first and second parameters indicate the paths to the original and compared images. The third (optional) specifies the path to the metric image {PNG}.

stbnhwmetrics ${IMAGE_ORIG_PATH} ${IMAGE_COMPARE_PATH} [metric.out.png]

structure

dependencies.c - API stb
metricsnhw.h - NHW metrics
metricspsnr.h - metrics PSNR and MSE
metricsum.h - "Universal Scale" of metrics (UM)
smallfry.h - metrics SMALLFRY, SHARPENBAD and Correlation
ssim.h - metrics SSIM and VIFP1
stb/ - stb
stbimmetrics.c - CLI program.
ycbcr.h - YCbCr color space.

Metrics

"Universal Scale" of metrics (UM):

  0.0
  ... (DIRTY) ...
  0.5
  ... (LOW) ...
  0.75
  ... (MEDIUM) ...
  0.875
  ... (SUBHIGH) ...
  0.9375
  ... (HIGH) ...
  0.96875
  ... (VERYHIGH) ...
  1.0

Trends:

  UM = 0.0713 * sqrt(sqrt(1.0 / MSE))
  UM = 0.109 * sqrt(PNSR)
  UM = 0.122 * sqrt(SDNSR)
  UM = 1.0 * cor_sigma(cor_sigma(COR))
  UM = 2.18 * cor_sigma(cor_sigma(SSIM))
  UM = 3.61 * cor_sigma(VIFP1)
  UM = 3.0 * (SMALLFRY / 100 - 0.5) * (SMALLFRY / 100 - 0.5)
  UM = (1.0 + 0.75 / radius) * cor_sigma(SHARPENBAD)
  UM = 0.261 * sqrt(sqrt(1.0 / NHW-C))
  UM = 0.286 * sqrt(sqrt(1.0 / NHW-N))
  UM = 0.223 * sqrt(sqrt(1.0 / NHW-R))

  cor_sigma(M) = 1.0 - sqrt(1.0 - M * M)

Origin and Compare:

Metrcics:

stbimmetrics lena.png lena.quant444.png lena.quant444.psnr.png 
Load: lena.png
image: 256x256:3
Load: lena.quant444.png
image: 256x256:3
type: psnr
metric: 18.663233
Save png: lena.quant444.psnr.png

MSE: 0.013604, UM: 0.208771
PSNR: 18.663233, UM: 0.470890
SDSNR: 13.888971, UM: 0.454668
Corelation: 0.953666, UM: 0.285007
SSIM: 0.618509, UM: 0.050606
VIFP1: 0.170903, UM: 0.053111
SMALLFRY: 83.753342, UM: 0.341786
SHARPENBAD: 0.389942, UM: 0.138531
NHW-N: 0.072673, UM: 0.550836
NHW-C: 0.145290, UM: 0.422748
NHW-R: 0.116387, UM: 0.381794