COMPARISON OF OPEN AND FREE

VIDEO COMPRESSION SYSTEMS

A Performance Evaluation

Till Halbach

Norwegian Computing Center, Norway

Keywords:

Still-image and video compression, Performance comparison, Dirac, Theora, H.264, Motion JPEG2000.

Abstract:

This article gives a technical overview of two open and free video compression systems, Dirac and Theora I,

and evaluates the rate distortion performance and visual quality of these systems regarding lossy and lossless

compression, as well as intra-frame and inter-frame coding. The evaluation shows that there is a substantial

performance gap of Theora and Dirac when compared to H.264- and Motion JPEG2000-compliant reference

systems. However, an algorithm subset of Dirac, Dirac Pro, achieves a performance comparable to that of

Motion JPEG2000, and which can be less than one dB below the PSNR performance of H.264 with TV-size

and HD video material. It is further shown that the reference implementations of the codecs of concern still

have potential for efﬁciency improvements.

1 INTRODUCTION

The video compression community has recently been

witness to an interesting development concerning

royalty-free speciﬁcations. Both the British Broad-

casting Corporation (BBC) and the Xiph.org Foun-

dation have released video codecs which are open-

technology and open-source and free to use for the

public. They compete therefore with patented codecs

such as MPEG-x and H.26x which are, though spec-

iﬁed in international standards, typically subject to a

license fee, and not entirely open. While the lack of

royalty claims makes these speciﬁcations interesting

candidates for use by industry and individuals like-

wise, the technologies provided should offer a perfor-

mance close to or better than today’s state-of-the-art

standards to ensure widespread acceptance.

So far, a detailed performance assessment of both

Dirac and Theora relative to existing standards has

been lacking in the literature; a gap this article wants

to ﬁll. All involved technologies are presented be-

low with technical details. Dirac and Theora are then

compared to each other and to reference standards to

evaluate their performance. Conclusions are drawn at

the end of the article.

2 DIRAC

The BBC have been working on a family of general-

purpose video codecs with the umbrella name Dirac

for some time now. As a result, two major products

have been released to the public at the time of writing.

Dirac Pro, which is version 1.0 of the Dirac fam-

ily, is an intra-frame video codec which targets loss-

less or near lossless image processing with low la-

tency, suitable e.g. for studio and professional appli-

cations (BBC, 2008a). Dirac Pro has been submit-

ted to the Society of Motion Picture and Television

Engineers and is expected to become international

standard VC-2 (W3C, 2008), with Microsoft’s WM-9

being VC-1 (SMPTE, 2006). Version 2.1 of Dirac is

a super-set of Dirac Pro, including motion estimation/

motion compensation (ME/MC) (BBC, 2008b). It is

developed mainly with broadcasting and streaming in

mind, targeting application areas like TV, digital cin-

ema, and Internet. According to the Dirac Pro speciﬁ-

cation, there is the intention to extend VC-2 to include

the full Dirac speciﬁcation.

In particular for the end user it is of inter-

est that the BBC claim the standards suite to

be patent-/royalty-free and open-technology (Borer,

2005). The reference codec developed is licensed un-

der the Mozilla Public License version 1.1.

Halbach T. (2009).

COMPARISON OF OPEN AND FREE VIDEO COMPRESSION SYSTEMS - A Performance Evaluation.

In Proceedings of the First International Conference on Computer Imaging Theory and Applications, pages 74-80

DOI: 10.5220/0001809700740080

 SciTePress

2.1 Technology Overview

Like international video compression standards,

Dirac deﬁnes the byte stream and decoding proce-

dure only. However, for illustration purposes, the

discussions below are made with regard to standard-

compliant encoders.

Dirac Pro (BBC, 2008a) accepts RGB and YUV

video, with chrominance subsampling factors of

4:4:4, 4:2:2, and 4:2:0. There are no restrictions

on the bit resolution, and neither on frame size or

frame rate. Predeﬁned formats deﬁned in the refer-

ence implementation include picture resolutions from

below QCIF size to 4K Digital Cinema. The nature

of the original video can be both interlaced and pro-

gressive. Dirac Pro does not involve any MC. A

compliant encoder comprises three main compression

components. A frame of a video sequence is ﬁrst

transformed by a wavelet ﬁlter bank to exploit spatial

redundancies, before the transform coefﬁcients are

quantized and entropy encoded to exploit statistical

correlation. The ﬁlter bank can be implemented in a

very efﬁcient way by means of lifting. The quantizers

used are basically scalar dead-zone quantizers and are

left out for lossless encoding. Entropy encoding can

either mean variable-length coding (VLC) or arith-

metic coding (AC). Concerning data migration, later

transcoding without loss is possible by compressing

video sequences in Dirac Pro’s lossless mode.

Another feature of Dirac Pro is the use of inde-

pendent chunks of data inside the byte stream, which

allows for fast forward and backward functionality.

All data are stored in a single byte stream, such that

the transport in container formats like MPEG trans-

port stream or OGG is possible. Due to the use of a

ﬁlter bank hierarchy in the decomposition stage, in-

cremental frame decoding is easily achieved, allow-

ing for scalability in quality and resolution. Temporal

scalability is supported implicitly as no temporal pre-

diction is involved.

Though Dirac Pro deﬁnes the from other stan-

dards well known concept of proﬁles (sets of algo-

rithms) and levels (constraints on decoder resources

like frame and data buffer sizes), none have been de-

ﬁned yet. However, a coding option for high quality

and another one for low latency exists. High quality is

achieved by processing the entire frame at a time in-

stance and implies thus a delay of at least one frame.

Low latency is achieved by processing subregions of

the frame at a time, and by replacing AC by VLC.

This comes typically at the expense of less efﬁcient

compression. By the use of a wavelet ﬁlter bank and

predicting subband coefﬁcients, Dirac Pro can be re-

lated to the well know SPIHT video compression al-

gorithm (Said and Pearlman, 1996) and the interna-

tional standard Motion JPEG2000 (ISO/IEC, 2001).

As mentioned before, Dirac 2.1.0 (BBC, 2008b)

deploys MC to exploit temporal redundancies and can

therefore be counted as a member of the family of

block-based hybrid video codecs. Dirac is identical

to Dirac Pro, except that here not the original image

is passed to the ﬁlter bank, but a motion compensated

one. As a consequence, the functionality of scalability

is lost. The motion prediction process is quite ﬂexible,

allowing any combination of various unidirectional

and bidirectional modes, including weighting. An ad-

ditional option for global motion estimation exists.

The block sizes utilized in the prediction process can

be of varying size and overlap each other, also known

as overlapped block motion compensation. The accu-

racy can be either half-pel, quarter-pel, or 1/8-pel.

3 THEORA

Theora I is a video codec which is developed by mem-

bers of the Xiph.org Foundation. It is more and

more frequently found on the Web and used by large

sites complying with the Wikimedia Commons like

Wikipedia. However, the main application areas re-

main undeﬁned. Xiph.org claims that Theora con-

tains solely royalty-free and open technology. Its ref-

erence implementation is provided under a BSD-style

license.

3.1 Technology Overview

Also this codec can be classiﬁed as belonging to the

family of block-based hybrid video codecs. As with

international standards, the speciﬁcation details the

decoder operation and data ﬁeld order in the byte

stream. As a concatenation of data packets, this

stream can easily be encapsulated in any suitable

transport container format. The standard allows pro-

gressive video material with 8bpp accuracy and arbi-

trary dimensions, ranging from below QCIF to sig-

niﬁcantly more than 4K. The sequence may have a

YUV color space with 4:2:0, 4:2:2, and 4:4:4 chroma

subsampling.

A compliant decoder comprises ﬁve main compo-

nents. The byte stream data are ﬁrst entropy decoded

employing Huffman codes. The decoded transform

coefﬁcients are then passed to the inverse quantization

process and, subsequently, to an inverse 8 × 8 DCT,

producing a frame difference signal, or frame delta.

COMPARISON OF OPEN AND FREE VIDEO COMPRESSION SYSTEMS - A Performance Evaluation

The frame delta pixels are then added to the predicted

frame to form the reconstructed frame which is pro-

cessed by a deblocking ﬁlter before picture display

and storage in the frame buffer. The ﬁlter has hence

an in-loop position, and it is applied to block edges.

The block-based MC process utilizes forward pre-

diction (P-frames) based on a single reference frame,

as well as simple block copying, but no bi-predictive

(B-) frames. The accuracy is either full pel or half pel.

Otherwise it is of interest that that most algorithms

make use of new data scanning orders in order not to

infringe any patents. Arbitrarily accessing I-frames in

the code stream is possible under the constraint that

the byte stream header must have been transmitted

and processed for decoder initialization previously.

Other than that, it is worth mentioning that Theora

does not provide the possibility for scalable picture

decoding, and it lacks further the option of later loss-

less data migration from data stored in Theora format

to another format, as it does not provide any means

for lossless coding.

4 PERFORMANCE

COMPARISON

In this section, Dirac and Theora are compared to

each other performance-wise, and to Motion JPEG-

2000 and H.264 as reference codecs and international

state-of-the-art video compression standards (ISO/

IEC, 2001; ITU-T, 2003). To the author’s knowledge,

only a single independent evaluation of Dirac has

been undertaken previously (Onthriar et al., 2006),

where the authors found that H.264 outperforms the

at that time emerging Dirac both in PSNR and SSIM

value, especially at low bot rates. Theora remains

untested in that respect.

All encoders are operated in rate distortion

optimization (RDO) mode or near RDO mode (if pro-

vided) and identical or near identical parameter sets

required for fair codec comparisons. The involved

motion compensation implies 250 P- or B-frames be-

tween I-frames and a single pass.

To assess the quality of Dirac 2.1.0, the version

0.9.1 of the reference implementation is used, which

complies with the standard. The encoder is quality

controlled by setting a quality factor similar to that

of the JPEG standard. Near RDO is accomplished

by requiring a full ME search with an area of 32 × 32

pixels. Concerning Theora I, version 0.19 of the refer-

ence implementation with the name libtheora is used.

The software is quality controlled like Dirac with a

quality parameter and is operated in RDO mode. For

compressing video material compliant with H.264,

the implementation named x264 is used. The encoder

is rate controlled but operated without any particular

RDO due to a lack of such an option. Two B-frames

are inserted between two P-frames, for which the ME

process is based on a single reference frame. The

search area is of the size 32 square pixels and has a

hexagonal form.

All video sequences used in the experiment com-

prise 8 bpp image material and a YUV color space

with 4:2:0 subsampling. AKIYO is a 300-frame head-

and-shoulder sequence in QCIF resolution with little

motion, MOBILE has SD/PAL resolution and consists

of 220 frames with a moderate amount of motion,

while CREW comprises HD 720p material with ﬂash-

ing effects.

4.1 Inter-frame Comparison

The comparison of inter-frame coding schemes treats

lossy and lossless coding separately.

4.1.1 Lossy Coding

The original video sequence is encoded and decoded

involving temporal prediction, and the average lumi-

nance PSNR (Y-PSNR, in dB) is recorded, as well as

the achieved bit rate in Kbps. The rate distortion (RD)

curves are drawn over six different codec operation

points, while only ﬁve points are recorded with The-

ora due to its limited allowed range for the quality

factor. The bit rate is measured for raw video data

except Theora, with which the data is encapsulated

inside the OGG container format, implying the use of

(a negligible amount of) overhead data. The size of

the video data is naturally somewhat lower.

50 100 150 200 250 300 350

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac

Theora

H.264

Figure 1: RD comparison for AKIYO (QCIF).

IMAGAPP 2009 - International Conference on Imaging Theory and Applications

It can easily be seen in Fig. 1 through Fig. 3 that

there are signiﬁcant performance gaps among all in-

volved codecs. The reference system H.264 outper-

forms both Dirac and Theora by a substantial amount,

in terms of PSNR, at all rates. The difference between

Dirac and H.264 can be more than 10 dB at low rates

and up to 5 dB at moderate to high rates. Dirac’s tar-

get application range is obviously high bit rate. Be-

sides the aforementioned, Fig. 3 also bears proof of

the fact that Dirac provides only a few coarse quan-

tizers, which limits its RD curve at low RD products.

Dirac also copes better with SD- and HD-sized video

material than with smaller formats. This can be de-

rived from the fact that the PSNR difference between

Dirac and H.264 is signiﬁcantly less (roughly 3 dB) in

Fig. 2 and Fig. 3 than in Fig. 1.

Theora performs worse than Dirac. In Fig. 1 and

Fig. 3, the reference implementation does not gain an

improved performance with a bit rate increase, which

is in contrast to what would have been expected intu-

itively. This is, however, likely an implementation is-

sue, as the codec performs as expected with MOBILE.

In either case, it can be concluded that neither The-

ora nor Dirac are suitable codecs for video sequences

with less-than-SD resolution, considering PSNR val-

ues below 30 dB at low bit rates. There are manifold

reasons for H.264’s superiority, and its hierarchical

ME mechanism with variable block sizes and a very

accurate interpolation, the adaptive in-loop deblock-

ing ﬁlter, as well as the highly efﬁcient arithmetic en-

coder are among them.

0 5000 10000 15000 20000 25000 30000

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac

Theora

H.264

Figure 2: RD comparison for MOBILE (SD).

The gap in PSNR between Theora and H.264 can

be as large as 10 dB, when the malfunctioning of the

implementation is ignored. Fig. 2 shows that the gap

decreases with roughly 2dB towards low bit rates.

Theora is obviously designed for low to moderate

bit rates. The container overhead is measured to be

around 1% and is hence without inﬂuence on the ﬁnal

result.

0 5000 10000 15000 20000 25000 30000 35000

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac

Theora

H.264

Figure 3: RD comparison for CREW (HD).

The efﬁciency of the involved implementations is

measured in Tab. 1. The respective software is in-

voked 20 times on a single general-purpose CPU with

2.2GHz clock frequency, and the CPU time needed

for encoding (in seconds) is averaged over all runs.

This comparison has to be interpreted with care, as

implementations like the one for Dirac are for a proof

of concept only and not optimized for speed. How-

ever, the comparison serves as a pointer for a rough

assessment of the encoders’ complexity.

Table 1: Implementation efﬁciency, in fps.

Dirac Theora H.264

Akiyo 21 379 323

Mobile 2 9 13

Crew 0.03 0.3 8

The Dirac implementation appears to be the most

inefﬁcient software of the three measured implemen-

tations. Dirac’s frame processing frequency is lowest

in all cases, while the Theora and H.264 implementa-

tions are alternately best. However, it is of advantage

that the processing frequency of Dirac’s software is

almost constant at varying bit rates, or in other words,

its standard deviation σ

Dirac

is much smaller than 1.

The standard deviation σ

Theora

equals roughly 31, 1,

and 0.07 with AKIYO, MOBILE, and CREW, respec-

tively, and σ

H.264

is roughly equal to 49, 5, and 2,

respectively. Summarizing, the frame processing fre-

quency of Theora’s implementation varies most.

It is concluded that Theora and H.264, which

both deploy a transform instead of a ﬁlter bank like

Dirac, can be operated much faster than the latter

mentioned codec, despite the lifting structure as de-

scribed in the Dirac speciﬁcation. Dirac is close to

real-time performance with QCIF-size video, while

COMPARISON OF OPEN AND FREE VIDEO COMPRESSION SYSTEMS - A Performance Evaluation

Figure 4: Original (left) and decoded frame 100 of AKIYO. Middle left: Dirac; Middle right: Theora; Right: H.264. The

images may be slightly scaled due to space considerations.

Theora and H.264 are way above this requirement.

Neither of the mentioned codecs is capable of en-

coding in real time with SD-size video or larger (i.e.,

without additional implementation optimization and

without extra hardware). Concerning Dirac, it should

be mentioned that a high-performance implementa-

tion named Schr

odinger is being developed, and that

specialized hardware exists for coding of HD signals

in real time (Borer, 2007).

A visual comparison of all involved codecs at

a very low bit rate reveals signiﬁcant artifacts with

both Dirac and Theora. Fig. 4 shows a single de-

coded frame of the AKIYO video. All videos were

encoded as described previously. The bit rates of

Theora and H.264 are 17.8 Kbps and 17.0 Kbps, re-

spectively, while the for the reference implementation

lowest possible bit rate with Dirac is 67.2Kbps. It

can easily be seen that, despite the signiﬁcant higher

bit rate, Dirac is the codec which performs worst,

i.e. which introduces much more compression arti-

facts than the two competing codecs. The artifacts are

mainly ringing along edges and blurred areas, which

can be attributed to the use of a ﬁlter bank instead of

a transform as with Theora and H.264. With Theora,

the decoder outputs a slightly blurred picture when

compared to H.264, and also minor blocking can be

observed (in the women’s face). Dirac and Theora

yield 26.5 dB and 30.5 dB with the frame of con-

cern, respectively, whereas H.264 achieves a PSNR of

34.3dB. Here, the decoded picture is slightly blurred

compared to the original, but no blocking artifacts are

found, despite the approach of block-based MC and

a block-based transform. The explanation lies in the

good performance of H.264’s in-loop deblocking ﬁl-

ter, which obviously outperforms the ﬁlter employed

in Theora. Note ﬁnally that, while the visual com-

parison (exempliﬁed by a single frame) suggests that

Theora is superior to Dirac at very low bit rates, the

RD points in Fig. 1 are averaged over the entire se-

quence.

4.1.2 Lossless Coding

As already mentioned, Theora does not offer a loss-

less option. Concerning Dirac, it was found that the

reference implementation was not stable enough to

conduct these experiments, hence this topic remains

for future research.

4.2 Intra-frame Comparison

This section investigates the efﬁciency of Dirac Pro

and Theora when disregarding motion compensation,

i.e. in pure intra-frame coding mode. With Motion

JPEG2000 or short MJ2, an additional international

state-of-the-art standard is taken as reference system.

Lossy and lossless coding are again treated separately.

Version 1.3 of the implementation named Open-

JPEG has been used to encode/decode to and from

MJ2 format. Other implementations with differing ef-

ﬁciency exist (Pearson and Gill, 2005). The encoding

process generates a single quality layer, and four and

ﬁve decomposition levels for AKIYO and MOBILE/

MOBILE, respectively. The lossy scenario involves an

irreversible 9/7 wavelet ﬁlter bank, while the lossless

scenario demands reversible 5/3 integer wavelets.

4.2.1 Lossy Coding

The experiments in this section are as described in

Sec. 4.1.1, except that MC is turned off, i.e. the en-

coded bit stream consists solely of I-frames. Fig. 5

through Fig. 7 lack RD curves for Theora as its ref-

erence implementation is not capable of running in

intra-frame mode.

Again, H.264 outperforms the other codecs signif-

icantly at high bit rates, while the difference in PSNR

can be less than 1 dB at low rates, depending on the

video material. It is further observed that Dirac Pro

has roughly the same RD behavior as Motion JPEG-

2000. In fact, Dirac’s intra-frame coding appears to

be optimized for compressing TV and HD material at

low bit rates, given a PSNR close to that of H.264 with

IMAGAPP 2009 - International Conference on Imaging Theory and Applications

0 500 1000 1500 2000 2500 3000 3500

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac Pro

Motion JPEG2000

H.264

Figure 5: RD comparison in intra mode for AKIYO (QCIF).

0 10000 20000 30000 40000 50000 60000 70000

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac Pro

Motion JPEG2000

H.264

Figure 6: RD comparison in intra mode for MOBILE (SD).

MOBILE/CREW. This is in contrast to the inter-frame

RD behavior as seen in Fig. 1 but can be explained by

a different set of quantizers and differing statistical

signal properties.

The latency of the reference implementations is

measured as described in Sec. 4.1.1, and listed in

Tab. 2. It is found that the H.264 software is best with

a substantial margin in terms of frame processing rate,

0 50000 100000 150000 200000 250000

Y-PSNR / [dB]

Bit rate / [Kbits/s]

Dirac Pro

Motion JPEG2000

H.264

Figure 7: RD comparison in intra mode for CREW (HD).

Table 2: Implementation efﬁciency, in fps.

Dirac Pro Motion JPEG2000 H.264

Akiyo 128 62 266

Mobile 8 3 15

Crew 5 2 11

and the implementation of Motion JPEG2000 is the

most inefﬁcient of the three implementations of con-

cern. However, the latter one performs in the most

uniform manner with a σ

MJ2

less than or equal to 2.

The deviations σ

Dirac Pro

and σ

H.264

are 47 and 79 with

QCIF-size material, 4 and 7 with PAL material, and

2 and 3 with HD material, respectively. I.e., the per-

formance efﬁciency of H.264’s implementation varies

most.

4.2.2 Lossless Coding

Dirac Pro in lossless intra-coding mode is compared

to both reference codecs. Tab. 3 shows the achieved

bit rate of all coding schemes. Dirac Pro is outper-

formed by both reference codecs. H.264 appears to

draw advantage of its highly efﬁcient intra-frame pre-

diction mechanisms which are superior to those em-

ployed in Dirac Pro. Motion JPEG2000, which in turn

is based on JPEG2000, can rely on superior subband

prediction and a highly efﬁcient arithmetic coding en-

gine. As already mentioned, the ﬁlter banks of Dirac

Pro and MJ2 are set to be identical, so the arithmetic

encoder of Dirac Pro has potential for improvement

when compared to Motion JPEG2000. A side result

of the experiments is that Motion JPEG2000 appears

to be the best choice for professional studio applica-

tions.

Table 3: Comparison of lossless intra-frame coding

schemes. The rate is given in Mbps.

Dirac Pro H.264 Motion JPEG2000

Akiyo 4.8 3.6 3.1

Mobile 102.3 86.8 76.5

Crew 335.1 276.2 224.8

5 CONCLUSIONS

Two open and free codecs were presented in this arti-

cle. Their technical details were discussed, and a per-

formance evaluation was given. It is stressed that the

assessment is by no means exhaustive, but rather must

be interpreted as a starting point for further research.

The experiments showed that the reference sys-

tem H.264 outperforms the two codecs signiﬁcantly,

COMPARISON OF OPEN AND FREE VIDEO COMPRESSION SYSTEMS - A Performance Evaluation

both in the objective and the subjective/visual evalu-

ation. The low performance (in terms of PS NR) of

Dirac and Theora surprises, since both codecs have

been designed to compete with current cutting-edge

technology. While Dirac is suitable for compression

of SD/HD-sized video, its reference implementation

is the slowest of the three systems of concern. The

implementation of Theora comes closest to the imple-

mentation of H.264 in terms of efﬁciency. However,

this system achieves the lowest PSNR of all involved

codecs. Dirac Pro can compete with Motion JPEG-

2000 over the entire rate spectrum and comes with

less than 1dB difference in PSNR close to H.264 with

TV and HD material.

It is concluded that the algorithms of the two

open and free codecs need improvement to be able

to compete with state-of-the-art technology. With to-

day’s performance both standards can, however, be

compared to former video compression systems like

MPEG-2 and H.263+, as the performance gap be-

tween those standards and e.g. H.264 is well known

(Wiegand and Sullivan, 2007). The battle for the

most used video compression format on the Web will

hence continue, and the candidates still have to posi-

tion themselves in a better way, i.e. with improved RD

behavior and other functionality. In particular, The-

ora should offer options for lossless coding and intra-

frame coding, and Dirac should provide for compres-

sion at very low bit rates. At the same time, the encod-

ing/decoding libraries have to be optimized for speed

to increase the possibility for a wide acceptance of the

offered technologies.

It is ﬁnally stressed that the comparisons pre-

sented in this article are not only comparisons of spec-

iﬁcations, but also comparisons of encoder implemen-

tations. Though all standard-compliant, different en-

coders typically yield different performances. The re-

sults presented here can therefore not be generalized

to be valid for other implementations or even other

versions than those involved.

REFERENCES

BBC (2008a). Dirac Pro speciﬁcation, version 1.0.0.

BBC (2008b). Dirac speciﬁcation, version 2.1.0.

Borer, T. (2005). Open technology video compression for

production and post production. EBU Technical Re-

view.

Borer, T. (2007). Open technology video compression

for production and post production. In International

Broadcasting Convention, Amsterdam (Netherlands).

ISO/IEC (2001). JPEG2000 Image Coding System, Part 3.

ITU-T (2003). Advanced video coding for generic audiovi-

sual services. Recommendation H.264.

Onthriar, K., Loo, K. K., and Xue, Z. (2006). Perfor-

mance comparison of emerging dirac video codec

with h.264/av. In ICDT ’06: Proceedings of the Inter-

national Conference on Digital Telecommunications,

page 6, Washington (DC, USA). IEEE Computer So-

ciety.

Pearson, G. and Gill, M. (2005). An evaluation of motion

JPEG 2000 for video archiving. In Proc. IS&T Archiv-

ing, pages 237–243, Washington (DC, USA).

Said, A. and Pearlman, W. A. (1996). A new fast and ef-

ﬁcient image codec based on set partitioning in hier-

archical trees. IEEE Trans. Circuits, Syst. for Video

Technol., 6:243–250.

SMPTE (2006). VC-1 compressed video bitstream format

and decoding process.

W3C (2008). Video on the web, workshop report. Ac-

cessed: 2008-02-28.

Wiegand, T. and Sullivan, G. J. (2007). The H.264/AVC

video coding standard. IEEE Signal Processing Mag-

azine, pages 148–153.

IMAGAPP 2009 - International Conference on Imaging Theory and Applications