Jump to content

User:AZelentsov/sandbox

From Wikipedia, the free encyclopedia


Quality Measurement

[edit]

Different deinterlacing methods have different quality and speed characteristics.

Usually, to measure quality of deinterlacing method, the following approach is used:

  1. A set of progressive videos is composed
  2. All of these videos are interlaced
  3. Each of interlaced videos are deinterlaced with specific deinterlacing method
  4. All of deinterlaced videos are compared with the corresponding source video via objective video quality metric, such as PSNR, SSIM or VMAF.

The main speed measurement metric is frames per second (FPS) - how many frames deinterlacer is able to process per second. Talking about FPS, it is necessary to specify the resolution of all frames and hardware characteristics, because the speed of specific deinterlacing method significantly depends on these two factors.

Benchmarks

[edit]

Deinterlacing Challenge 2019[1]

[edit]

This benchmark has compared 8 different deinterlacing methods on a synthetic video. There is a moving 3-dimensional Lissajous curve on the video in order to make it challenging for the modern deinterlacing methods. The authors used MSE and PSNR as objective metrics. Also, they measure processing speed in FPS. For some methods there is only visual comparison, for others - only objective.

Another algorithms of Deinterlacing Challenge 2019
Algorithm MSE PSNR Processing speed
(FPS)		 Open source
Vegas De-interlace Blend[2] 8.086 43.594 3.53 No
Vegas De-interlace Interpolate[3] 16.426 41.292 3.58 No

MSU Deinterlacer Benchmark[4]

[edit]

This benchmark has compared more than 20 methods on 40 video sequences. Total length of the sequences is 834 frames. Its authors state that the main feature of this benchmark is the comprehensive comparison of methods with visual comparison tools, performance plots and parameter tuning. Authors used PSNR and SSIM as objective metrics.

Top algorithms of MSU DIB
Algorithm PSNR SSIM Processing speed
(FPS)		 Open source
MSU Deinterlacer[5] 40.708 0.983 1.3 No
VapourSynth TDeintMod[6] 39.916 0.977 50.29 Yes
NNEDI[7] 39.625 0.978 1.91 Yes
FFmpeg Bob Weaver Deinterlacing Filter[8] 39.679 0.976 46.45 Yes
Vapoursynth EEDI3[9] 39.373 0.977 51.9 Yes
Real-Time Deep Video Deinterlacer[10] 39.203 0.976 0.27 Yes

VapourSynth TDeintMod author states that it is bi-directional motion adaptive deinterlacer. NNEDI method uses a Neural Network to deinterlace video sequences. FFmpeg Bob Weaver Deinterlacing Filter is the part of well-known framework for video and audio processing. Vapoursynth EEDI3 is the abbreviation for "enhanced edge directed interpolation 3", authors of this method state that it works by finding the best non-decreasing warping between two lines according to a cost functional. The authors of Real-Time Deep Video Deinterlacer use Deep CNN to get the best quality of output video.

See also

[edit]

References

[edit]
  1. ^ https://www.vegascreativesoftware.info/us/forum/deinterlacing-challenge-2019--114485/
  2. ^ https://www.vegascreativesoftware.com
  3. ^ https://www.vegascreativesoftware.com
  4. ^ https://videoprocessing.ml/benchmarks/deinterlacer.html
  5. ^ https://compression.ru/video/deinterlacer/index.html
  6. ^ https://github.com/HomeOfVapourSynthEvolution/VapourSynth-TDeintMod
  7. ^ https://github.com/dubhater/vapoursynth-nnedi3
  8. ^ https://ffmpeg.org/ffmpeg-filters.html#bwdif
  9. ^ https://github.com/HomeOfVapourSynthEvolution/VapourSynth-EEDI3
  10. ^ https://arxiv.org/pdf/1708.00187.pdf
[edit]

Category:Video processing

Retrieved from "https://en.wikipedia.org/w/index.php?title=User:AZelentsov/sandbox&oldid=1015480605"