Coding arbitrary-shape objects with MPEG-4

[xyiann00]

Hi

I have read that MPEG-4 supports coding arbitrary-shape
objects (so called VOP's) using different coding parameters for each object.

I have a scene with a narrator and a blue background and would
like to code the backround using lower resolution.

Do you know of any encoder that supports this function?

From what I understand the object segmentation process
is not part of the encoder but a pre-processing step.
I understand I have to define the objects using binary
alpha mask. Is there a syntax language that the MPEG-4
encoder would understand? How do i multiplex the two VOP's
into a single bitsream?

I would appriaciate any help. Thanks.

[Sirber]

maybe Luma masking could quant more your blue background :)

[hanfrunz]

i think there is no software codec that uses VOPs. Maybe some special harware for videoconferencing is using it.

[xyiann00]

I think any MPEG-4 compliant codec can automatically encode arbitrary-shaped objects but I am not absolutely sure about it.

The problem is to define the objects, which is done "outside"
the encoder perhaps with the help of the MPEG-4 Syntactic Description Language (MSDL). But still I don't know how to interface with the encoder.

Is it possible to multiplex one video sequence with two or more
audio sequences (for example one for every language) into one bitstream?

I think it can be done inside an avi container but don't know about the mp4. If it can be done with audio why not with two
video sequences that overlap each other?

I am a bit confused, so I am sorry if this does not make sense.

[BoNz1]

http://fame.sourceforge.net/ I have no idea if it works or how well it works but it does support arbitary shape.

[bond]

http://fame.sourceforge.net/ I have no idea if it works or how well it works but it does support arbitary shape.funky, what profile is this part of?

[BoNz1]

funky, what profile is this part of?

http://www.m4if.org/resources/mpeg4userfaq.php#ArbitraryObjects

Cheers.

[xyiann00]

The following visual profiles support arbitrary shaped objects

Core
Main
N-bit
Advanced Coding Efficiency
Core Scalable

The following visual profiles do not support arbitrary shaped objects

Simple
Simple Scalable
Advanced Real Time Simple
Advanced Scalable Texture
Advanced Simple
Fine Granularity Scalable

Do you know of any encoder that supports these Profiles?

The famelib project does support arbitrary shaped objects
but it is in a state not suitable for less advanced users like myself. I don't know how to use it.

[bond]

thx for the info

are there any tools which fame can be used in? both playback and encoding related?

[Valeron]

imo, currently there's no implementation can encode arbitrary shape in image compression.

despite the real compression saving compare to the mainstream tech, the efficiency-shape-identification is some kind of AI(if this come true, encoding will be much more mathmatics intensive) matter. assume no one care the encoding duration or we've got 8 core Intel Conroe IV(maybe a never available model:D ), the more efficiency coding tool introduce in the future may show the shape encoding is not that "needed" to be developped.

imo the shape-identification tool should be develop for image post processing to lower the manpower requirement.

[Manao]

despite the real compression saving compare to the mainstream techHuh ? It doesn't save anything in MPEG4, it actually decreases the coding efficiency. In AVC, one might be able to create pathological pictures where the use of arbitrary shape coding would improve marginally the coding efficiency because it'd help make contexts more coherent. But I'm not even sure such picture could be created.