MSU Brightness Independent PSNR (BI-PSNR)
- Project, Ideas: Dr. Dmitriy Vatolin
- Implementation: Alexey Noskov
- Updating and additions: Sergey Grishin
Common idea
BI-PSNR metric is intended for measuring distortions in video taking into account brightness shifts.
Change Log
[!] — Known bug
[+] — New Feature
[*] — Other
Version 1.1
[*] Visualization bug fixed for non-stadard resolution video
Version 1.0
[+] First plugin release
Usage
BI-PSNR metric should be used when one of the sequences has any brightness transformation, which does not change within frame. Example of such transformation is uniform increasing of brightness of contrast for single frame of for all sequence. Such transformations prevent usage of standard metrics because of strong brightness difference between comparing frames. BI-PSNR algorithm calculates brightness transformation, which makes frames similar as possible and calculates standard PSNR and MSE metrics taking into account founded transformation.
Visualization
There are two part of visualization:
- MSE visualization for frame. Colors of visualization are standard for MSE (in order of error increasing): black-blue-green-red.
- Brightness transformation plot. X-axis is brightness at the first sequence, Y-axis - brightness on the second one. Green points are values, which corresponds to each other (brightness transformation). Red diagonal is identical transformation (no brightness changes).

Example of visualization

Visualization of the same frame using standard PSNR
Plots
Plots of per-frame PSNR values after the found transformation are drawing after all calculations. Plots are entirely the same as standard per-frame PSNR

Plot's example

Standard PSNR plot for the same sequence
Algorithm
Table C[i,j] is filling for each frame: C[i,j] = { number of points in the same position, which have brightness i at the first sequence frame and j at the second sequence frame }
Next, for each i (brightness value from the first frame) we find corresponding brightness from the second sequence. Following formula is used to estimate distanse from arbitrary values of i and j:
One can note that this formula is sum of quadratic differences between all pixels of the first sequence with value i and all corresponding pixels from second sequence on the assumption that brightness was shifted to i-j.
When transformation was found, we can find MSE for the frame taking into account this transformation:
Download
- Metric’s sources for VS 2013
- MSU Video Quality Measurement Tool
- License text for sources and binary files (LGPL)
Contacts
E-mail: video-measure@graphics.cs.msu.ru
-
MSU Benchmark Collection
- MSU Super-Resolution for Video Compression Benchmark 2022
- MSU Video Quality Metrics Benchmark 2022
- MSU Video Upscalers Benchmark 2022
- MSU Video Alignment and Retrieval Benchmark
- MSU Mobile Video Codecs Benchmark 2021
- MSU Video Super-Resolution Benchmark
- MSU Shot Boundary Detection Benchmark 2020
- MSU Deinterlacer Benchmark
- The VideoMatting Project
- Video Completion
- Codecs Comparisons & Optimization
- VQMT
- Video Quality Measurement Tool 3D
-
MSU Datasets Collection
-
Metrics Research
- Video Filters
-
Other Projects