Документ взят из кэша поисковой машины. Адрес оригинального документа : http://graphics.cs.msu.ru/ru/science/research/videoqualityincreasing/deblock
Дата изменения: Sun Apr 10 02:20:24 2016
Дата индексирования: Sun Apr 10 02:20:25 2016
Кодировка: UTF-8
Removing of blocking effect from video | Лаборатория компьютерной графики и мультимедиа

Removing of blocking effect from video

MSU Graphics & Media Lab (Video Group)

The filter is intended for recovering visual quality of video ripped from DVD (for example, when it contains 4 hours of video data), VideoCD or after decompressing by H.261, H.263, H.264, DivX, XviD, etc. The algorithm decreases blocking artifacts, significantly improving visual quality of video.

Filter automatically determines the blockiness's strength on the frame and in the certain part of it, preserving the maximum of details. Thus, filter will process marginally when on scenes with slow motion, while it will smooth very hard in the case of fast motion in the same film.

The main advantage of the filter is that it raise the integral quality in the most widespread PSNR metric in comparison with original film. That is, a film after decompression in comparison with original has lower objective quality than a film after decompression and deblocking. (But for all that the original film is not used nowise, only decompressed data is used.) The filter can be easily implemented in hardware.

Examples

Filter's performance can be easily analized using the samples bellow.

Algorithm automatically determines blockiness's strength in each block border pixel of the frame and performs the processing accordingly. Thus, it uses hard smoothing in areas with strong blockiness while do nothing in areas with high contrast textures. Filter performance for case of hard blockiness is shown on picture bellow (frame fragment from sequnce "Battle" - the part of the "Terminator 2" movie):

Frame before processing The same frame after processing

Next picture (sequence "foreman", frame 152, compressed using Xvid with bitrate setting equal 50kbs) is an exmaple of mixed area containing both areas with contrast texture ("SIEMENS" inscription) and areas with middle-strength blockiness. It can be easily seen that blockiness is entirely suppressed together with no any degradation visible in inscription area:

 

Frame before processing The same frame after processing

Pictures series bellow illustrates visual performance of deblocking algorithm depending on the bitrate of compressed video. It can be easily seen that the more the strength of the blockiness present the higher radius smoothing is used for removing that blockiness. But at the same time all edges and contrast areas aren't oversmoothed. Pictures are located vertically in order of increasing bitrate (sequence "Foreman", frame 155, compressed using Xvid with bitrate settings: 50, 75 and 100kbs):

There are two branches on each diagram below. Branch "After decoder" corresponds to sequence right after the decoder without any post-processing. Second branch ("After decoder & MSU deblocking") corresponds to decompressed sequence post-processed using MSU deblocking. Vertical direction of the diagrams is graduated using Y-PSNR metric between corresponding and original sequences. Horizontal direction conforms to the bitrate (kbit/sec). These diagrams demonstrate that the quality of sequences after deblocking is significantly higher comparing to decompressed sequences. The Y-PSNR gain after performing of MSU deblocking achieves 1dB at low bitrates!



Publications

D.Vatolin, S.Grishin, "Method for High Quality Deblocking of Video without use of Quantization Infromation", Graphicon-2004. (Russian)

Download

Filter for VirtualDub MSU Deblocking (68 KB, ZIP)
Unpack this file in the folder "Plugins" in VirtualDub directory to use.

See also

Smart deblocking algorithm
Detailed description of MSU Deblocking filter
MSU SmartDeblocking v0.1 filter page (new deblocking algorithm)

The project team

Project lead:

  • Dr. Dmitriy Vatolin

Researchers:

  • Sergey Grishin

Contacts

e-mail:
video@graphics.cs.msu.ru