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

For professional users and codec authors

Part3. Additional testing

Contents

These diagrams show average shift of brightness produced by the codecs. Coordinates of the basic points represent average brightness values for the whole sequence and the number of measurement (bit rate). So each branch contains ten points that correspond to different bit rates.

Bankomatdi sequence

Picture 36.    Y-difference. Sequence bankomatdi

Conclusions:

• All the H.264 codecs raise average brightness in the sequence. The more is the bit rate the more brightness changes.
• Fraunhofer codec raises brightness stronger than the other codecs.

Battle sequence

Picture 37.    Y-difference. Sequence battle

Conclusions:

• All the H.264 codecs raise average brightness in the sequence. The more is the bit rate the more brightness changes.
• Videosoft codec raises brightness stronger than the other codecs.

Bbc3di sequence

Picture 38.    Y-difference. Sequence bbc3di

Conclusions:

• All the H.264 codecs raise average brightness in the sequence on the average and high bit rates. The more is the bit rate the more brightness changes.
• Change of brightness on the low bit rates is small.

Foreman sequence

Picture 39.    Y-difference. Sequence foreman

Conclusions:

• All the H.264 codecs except Fraunhofer strongly raise average brightness in the sequence. The more is the bit rate the more brightness changes.

Susidi sequence

Picture 40.    Y-difference. Sequence susidi

Conclusions:

• All the H.264 codecs strongly raise average brightness in the sequence. This change is approximately constant for all the bit rates.

These diagrams show how codecs handle different bit rates. Every branch consists of ten points that correspond to different bit rates. Value "1" means that the codec keeps the specified bit rate; value more than "1" means that the codec actually compresses with the higher bit rate than was specified.

Bankomatdi sequence

Picture 41.    Bit rate handling. Sequence bankomatdi

Conclusions:

• Videosoft codec slightly increases specified low bit rates.
• DivX codec increases specified low bit rates.

Battle sequence

Picture 42.    Bit rate handling. Sequence battle

Conclusions:

• Videosoft codec slightly increases specified low bit rates.
• DivX and MainConcept codecs strongly increase specified low bit rates.

Bbc3di sequence

Picture 43.    Bit rate handling. Sequence bbc3di

Conclusions:

• All the H.264 codecs slightly increase specified low bit rates.
• DivX codec strongly increases specified low bit rates.

Foreman sequence

Picture 44.    Bit rate handling. Sequence foreman

Conclusions:

• MainConcept significantly decreases specified bit rates.
• DivX increases specified bit rates.

Susidi sequence

Picture 45.    Bit rate handling. Sequence susidi

Conclusions:

• DivX increases specified low bit rates.

These diagrams show how compression quality changes from frame to frame. X-axis coordinate represents the number of the frame and Y-axis coordinate represents PSNR-value for this frame. Significant constant difference on the susidi diagram is caused by the noise suppression in every second frame (B-frames options in MPEG-2 coder).

Bankomatdi sequence

Bit rate = 100 Kb/sec

Bit rate = 700 Kb/sec

Bit rate = 2340 Kb/sec

Battle sequence

Bit rate = 100 Kb/sec

Bit rate = 700 Kb/sec

Bit rate = 2340 Kb/sec

Bbc3di sequence

Bit rate = 100 Kb/sec

Bit rate = 700 Kb/sec

Bit rate = 2340 Kb/sec

Foreman sequence

Bit rate = 100 Kb/sec

Bit rate = 700 Kb/sec

Bit rate = 2340 Kb/sec

Susidi sequence

Bit rate = 100 Kb/sec

Bit rate = 700 Kb/sec

Bit rate = 2340 Kb/sec

• Comparison was performed between the Ateme and DivX codecs.
• Bit rate was set to 700 Kb/sec.
• Bbc2di and foreman video sequences were used.

Bbc3di sequence, frame 280

The following examples demonstrate better image quality after compression with the H.264 codec in comparison with DivX.

Foreman sequence, frames 282 and 9

The following examples demonstrate better image quality after compression with the H.264 codec in comparison with DivX.

Conclusion:

• On similar PSNR values H.264 standard codecs show significantly better visual quality than previous standard codecs.
• Most of tested codecs are optimized for maximum speed on current day computers and do not implement all possibilities of H.264 format

The following examples demonstrate close results from the image quality point of view for DivX and H.264.