First Annual MSU MPEG-4 AVC/H.264 Video Codec Comparison

For professional users and codec authors

Return to first H.264/AVC comparison home page!

Part 1. Introduction

Contents

• Introduction
• PSNR/Bitrate diagrams
• Additional testing
• Final part

Thanks

We would like to thank Moonlight Cordless LTD, Fraunhofer Institute for Integrated Circuits IIS and Ateme for kindly providing us their codecs for this test.

Overview

Codecs

CODEC	MANUFACTURER	VERSION
1. Mpegable AVC Codec	dicas digital image coding GmbH	0.10
2. Moonlight H.264 Video Codec	Moonlight Cordless LTD	0.1.2546
3. MainConcept H.264 Codec	MainConcept AG	1.04.02.00
4. MPEG-4 / AVC Codec	Fraunhofer Institute for Integrated Circuits IIS	Date 25.11.2004
5. Ateme MPEG-4 AVC / H.264 Codec	Ateme	1.0.3.2
6. Videosoft H.264 Codec main	Videosoft, Inc	2.1.0.2
7. DivX Pro™ 5.1.1 Codec	DivXNetworks, Inc	5.1.1

Video sequences

Sequence	Number of frames	Number of frames per second	Resolution and color space
1. bankomatdi	376	30	704x352(RGB)
2. battle	1599	24	704x288(RGB)
3. bbc3di	374	25	704x576(RGB)
4. foreman	300	15	352x288(RGB)
5. susidi	374	25	704x576(RGB)

Goal and testing rules

Goal of H.264 codecs comparison

The main goal of this work was the comparative assessment of codecs quality in terms of consumer video compression. The video sequences used to perform this assessment were processed with a simple prevalent deinterlacing filter. The codecs were used with their default options.

Testing rules

• PSNR was measured using luv_avi program.
• Frame size was calculated as the ratio of the sequence size to the number of frames.
• Y-axis values on the Delta diagram were calculated as the difference between PSNR of the tested codecs and PSNR of the DivX codec.
• To test the codecs that place their logo on the frames of the compressed sequence a black rectangle was placed in its position on the both source and compressed sequences before their comparison.
• Compression with VfW (Video for Windows) codecs was performed using VirtualDub 1.5.4 video processing tool.
• Compression with DirectShow codecs was performed using GraphEdit (build 011008) tool.
• Compression with the codecs that had been installed as the separate applications was performed using these applications.
• For those codecs that compress video into their own internal format instead of avi the resulting compressed sequences were transformed into avi using GraphEdit (build 011008) and decoder given with the codec.
• MainConcept codec inserted extra frames into decoded sequences. These frames were deleted using VirtualDub 1.5.4 to perform per frames comparison. If the last frame in the corrected sequence differed visually from the last frame in the source uncompressed sequence comparison was not performed for this sequence.

Sequences

Bankomatdi

Sequence title	bankomatdi
Resolution	704x352
Number of frames	376
Color space	RGB
Frames per second	30
Source	MPEG-2 (DVD), Smart Deinterlace

This sequence is a fragment of the “Terminator-2” movie, which represents the scene near the cash dispenser. The sequence is characterized by slow motion, very little change of background (in the second part of the sequence camera slowly goes to the right) and comparatively high resolution.

Battle

Sequence title	battle
Resolution	704x288
Number of frames	1599
Color space	RGB
Frames per second	24
Source	MPEG-2 (DVD), FlaskMPEG deinterlace

This sequence is also a fragment of the “Terminator-2” movie, which represents the very beginning of the film. In the terms of compression this sequence is the most difficult one among all other sequences that took part in the testing. It is caused by three main reasons: constant changing of brightness because of the explosions and laser flashes, very quick motion and frequent changes of the scene that make codecs often compress frames as I-frames.

Bbc3di

Sequence title	bbc3di
Resolution	704x576
Number of frames	374
Color space	RGB
Frames per second	25
Source	Original (standard sequence), Smart Deinterlace

Picture 3. Frame 185 from bbc3di sequence

Picture 4. Frame 258 from bbc3di sequence

This sequence is characterized by pronounced rotary motion. It contains a rotating striped drum with different pictures and photos on it. Quality of the compressed sequence can be estimated by the details on these images.

Foreman

Sequence title	foreman
Resolution	352x288
Number of frames	300
Color space	RGB
Frames per second	15
Source	Original (standard sequence), progressive

Picture 5. Frame 77 from foreman sequence

Picture 6. Frame 258 from foreman sequence

This is another standard sequence. It represents a face with very rich mimic. On the one hand motion here is not very intensive, but on the other it is disordered, not forward. Intricate character of motion creates problems for the motion compensation process. In addition camera is shaking which makes the image unsteady. In the end of the sequence camera suddenly turns to the building site and there follows an almost motionless scene. So this sequence also shows codec’s behavior on a static scene after intensive motion.

Susidi

Sequence title	susidi
Resolution	704x576
Number of frames	374
Color space	RGB
Frames per second	25
Source	MPEG-2 (40Mbit), Smart Deinterlace

This sequence is characterized by high-level noise and slow motion. In its first part the scene is almost static (the girl only blinks), then there is some motion (she abruptly moves her head) and then the scene becomes almost static again. Noise is suppressed on every second frame due to the B-frames option in MPEG-2 codec.

Codecs

Mpegable AVC Codec

• This is a VfW (Video for Windows) codec.
• Compression was performed using VirtualDub 1.5.4 video processing tool.
• The codec is freeware.

Remarks:

VirtualDub jobs support contains an error on different bit rates. When trying to create several jobs only the last one is performed. Possible source is that codec might not save its settings during job creation.

Moonlight H.264 Video Codec

• This is a DirectShow codec.
• Compression was performed using GraphEdit (build 011008) tool.
• The version of the codec used in the testing was kindly provided by Moonlight Cordless LTD company.

Remarks:

Codec was tested without problems.

MainConcept H.264 Codec

• This codec is provided a separate application for video compression.
• The codec installs its decoder to the system therefore the compressed video can be played with the standard player.
• The codec is freeware.
• GraphEdit (build 011008) tool was used for decoding.

Remarks:

• The codec compresses only sequences with 30fps(NTSC) and 25fps(PAL) frame rates.
• The codec doesn’t compress with bit rate lower than 200 (settings lower than that are simply replaced by 200 without informing a user about it).
• Extra frames were added to the sequence while decoding the compressed sequences. Generally the less the bit rate was the more often extra frames were inserted. At the same time we could not treat these frames as drop-frames, because the resulting sequence length was increased. They had to be removed manually in order to perform per frame comparison.

Fraunhofer IIS Codec

• This codec is provided as a two separate applications for video compression and playback.
• The version of the codec used in the testing was kindly provided by Fraunhofer IIS company. This version works in command line.

Remarks:

• Only YUV format is acceptable as an input of console version, AVI format is acceptable for GUI version.

Ateme MPEG-4 AVC / H.264 Codec

• The version of the codec used in the testing was kindly provided by Ateme company
• This codec works with command line.
• The codec uses drop-frames to produce low bit rates.
• Input file should have YV12 format.

Remarks:

Codec was tested without problems.

Videosoft H.264 Codec

• This is a VfW (Video for Windows) codec.
• Compression was performed using VirtualDub 1.5.4 video processing tool.
• The codec is shareware with 5 days of trial use.

Remarks:

Codec was tested without problems.

DivX Pro™ 5.1.1 Codec

• This is a VfW (Video for Windows) codec.
• Compression was performed using VirtualDub 1.5.4 video processing tool.

Remarks:

Codec was tested without problems.