Features
Softage H.263 codec corresponds to the
Profile 0 of ITU-T Recommendation H.263 Video coding for low bit rate
communication. Coding mode and stream format of the Profile 0 are defined in ITU-T
H.263 Annex X Profiles and Levels Definition. Profile 0, also known as basic
profile, is supposed that there is no optional modes are implemented in Soft
Age release of H.263 codec.
The following distinctive features of the
implemented codec are to be mentioned:
Coding with fixed bitrate. Softage H.263 encoder
dynamically changes quality of several parts of coded frames in order to retain
approximately the same amount of bits in each encoded frame. |
Fast search algorithm of motion vectors for
effective interframe compression. |
There are no key frames in the compressed video
stream (except the only first one). Instead of using key frames Softage H.263
encoder make several parts of image to the key part in different periods of
time. Consequently, bitrate dependence of time is smoother as compared to
distinctive key frames with perfect quality of key parts. |
Compression is made per frame in contrast to
reference implementation. It allows more effective management of compression
parameters. |
Time critical algorithms are implemented using
Intel MMX technology. |
Since Intel MMX technology doesnt support
integer division, Softage H.263 encoder uses specially designed algorithm of
Fourier coefficients quantization which bypasses this limitation. Overall
performance of quantization is in six times faster then ordinary implementation
in pure C. |
Performance
All performance tests were made on
foreman 176x144 original video sequence. The figure below shows dependence of
encoding time on frame number in the sequence. One can see that the encoding
time lies between 4 and 5 milliseconds (on PC with Intel Pentium 4 CPU with 2
GHz frequency) except some sharp peaks. Average encoding time per frame is
about 5 ms and slightly varies with output video quality.

Foreman 176x144 original video sequence.

Performance of H.263 encoder.
Also quality of encoding was tested. As
distortion measure of whole sequence the averaged peak signal-to-noise ratio
(PSNR) was used, i.e. first PSNR for each frame n was calculated
according to
,
where M is the number of samples
in a frame, and oi and ci are the
amplitudes of the original and coded frame consequently. Then the PSNR values
of each frame are averaged for N frames in the sequence according to
.
All data presented on the figures below
were obtained for first 200 frames of foreman original video sequence in
176x144 resolution with using fixed quantizes 31, 24, 20, 16, 12, 10, 8 and 6.
Comparison with reference software was made by using data from paper Bernd
Girod et al. Comparison of H.263 and H.261 video compression standards
published in SPIE proceedings vol. CR60, Standards and Common Interfaces for
Video Information Systems, October 25-26, 1995, Philadelphia, USA.

Quality dependence of bitrate for H.263 codec.

Per frame PSNR dependence for Softage H.263 implementation for foreman sequence.
|