[フレーム]

ツ鞴褓 Video

ム?珞?褊?? ??蒟??? MSU Video Codecs Comparisons

Video Quality Measurement Tool (MSU VQMT)

Benchmarks

3D-?ソス?ソス?ソス?ソス?ソス 3D video

マ??裲??

Projects

ハ?鞳? «フ褪?蕘 ?赳??? 萵????»

ミ珸蒟??

About ホ ?褞粢??

News:

News:

RUSSIAN

Introduction to Video Codecs Comparison
MSU Graphics & Media Lab (Video Group)

Take a look at this article on the new site! Follow the link
https://videoprocessing.ai/codecs/comparison-introduction.html

Short introduction

There are a lot of different codec comparisons on the internet - long and short, emotional or not, biased and unbiased. But most of the authors of such comparisons just put two frames from 2 movies nearby, forgetting following important issues:

1. Every codec gives different quality on different frames of same sequence

Various factors play a role here. First, all codecs have a bitrate control mechanism, that produces quality fluctuations even in very good implementation. Second, bitrate control strategy is selected by user, and in CBR case on slow scenes quality will be low, and on fast ones it will degrade. Third, codecs usually have a so-called key frame, whose quality is determined separately and differs from other frames. Next, final quality depends on prefiltrations, that is included on all present-day codecs. That means that on every sufficiently long movie (average movie is about 150.000-200.000 frames) it is possible to select very good and very bad frames, especially if one-pass CBR compression was used, and movie had enough motion. As a result, having two codecs almost identical in terms of quality, you can easily show the superiority of either codec on specific frames. That method is widely used, alike in press-releases and video processing sites.


Diagram brings into accord with metrics PSNR (if higher, then better quality).
If you make a comparison on "peak" frames, the VP result will be better then DivX one.
Other frames show the opposite conclusion.

2. Different codecs are tuned to different film types.

It is known that some time ago simultaneously existed two different codecs - DivX 4.1 LowMotion and DivX 4.1 FastMotion. Playback was the same, but their internal parameters were tuned to low and high motion, and compression produced different results. Note - FORMAT was same in both cases, but compression was different, so equal size movies had different quality. Similar devotion to specific data type exists in all codecs. Some codecs better perform on high bitrates, while others better compress low ones. Some are tuned to high motion, and some better do with low. Codecs can be tuned to better compress animation or real video. Noisy and clear movies require separate tuning also. That means, that if you have almost equal codecs A and B, it is easy to find movies where A is better than B and vise versa. Curiously this could be same movie, but one could be taken from DVD and other re-filmed in cinema.

Original frame
Original frame Divx 3.1
Divx 3.1 Divx 5.1
Divx 5.1
This is frames quality in conventional colors (black - no difference with original,
blue and green - more difference, red - huge difference). If we will use this frames
for comparison - DivX 3.1 will be better DivX 5.1.

3. Compression quality could seriously depend on coding parameters.

Most people don't think about coding parameters, blaming it on the developers. Most codecs have many settings that can seriously affect quality First of all, these are bitrate strategy parameters, especially selection between "quality-based" and "bitrate-based" modes. After that prefiltration parameters play a role. For example, deinterlacing (because some movies were compressed with MPEG4 in interlaced mode), denoising (noise reduction), deflicking (removal of effect) and so on. There are parameters of key frame frequence, B-frame mask, control of prefiltering dependence from movie. etc. Here it is even easier to manipulate the results of a comparison of two codecs by simply optimizing the parameters of one of them. One can even put both movies on his site so everyone could see the difference (like some companies do) :-). In that case fact that one codec worked 10 times slower than another somehow becomes lost - why the people (and journalist) should to know such uninteresting information? In that case source sequence could not be found ( due to size, for sure), and coding parameters are unknown (and now you know why). Really there are many more delicate ways to receive "advantage" and even real advantage - for example there are no criteria to determine how frame skipping is significant for perception. But these methods are most common. So, if you study "fooling" methods, you could see that sufficiently large testing is required for correct comparison. As a minimum it is important to compare sequences with different characteristic on different bitrates, while using integral (average) quality characteristics throughout the sequence. So, let us wish you successful detailed comparison.

frame
Quality fall in one of the bitrates on this codec, because of the rate control error.
This bitrate is unlucky for this codec.
Good luck!

E-mail:
See all MSU Video Codecs Comparisons

MSU video codecs comparisons resources:

Other materials

Video resources:

3D and stereo video
Projects on 3D and stereo video processing and analysis
MSU Video Quality Measurement tools
Programs with different objective and subjective video quality metrics implementation Codecs comparisons
Objective and subjective quality evaluation
tests for video and image codecs
Ext. link: x264 parameters efficiency comparison
Public MSU video filters
Here are available VirtualDub and AviSynth filters. For a given type of digital video filtration we typically develop a family of different algorithms and implementations. Generally there are also versions optimized for PC and hardware implementations (ASIC/FPGA/DSP). These optimized versions can be licensed to companies. Please contact us for details via video(at)graphics.cs.msu_ru. Filters for companies
We are working with Intel, Samsung, RealNetworks and other companies on adapting our filters other video processing algorithms for specific video streams, applications and hardware like TV-sets, graphics cards, etc. Some of such projects are non-exclusive. Also we have internal researches. Please let us know via video(at)graphics.cs.msu_ru if you are interested in acquiring a license for such filters or making a custom R&D project on video processing, compression, computer vision.
Video codecs projects
Different research and development
projects on video codecs Other
Other information
Last updated: 28-May-2025
Server size: 8069 files, 1215Mb (Server statistics)

Project updated by
Server Team and MSU Video Group

Project sponsored by YUVsoft Corp.

Project supported by MSU Graphics & Media Lab

AltStyle によって変換されたページ (->オリジナル) /