Adapting algorithm for streaming environments

[MKZ]

The current design of the algorithm limits it to be used for compressing images individually. However, I'm interested in adapting this algorithm for streaming purposes i.e. define image size, color depth etc during initialization then just keep passing in image data (as memory pointer) and get encoded data out (as memory pointer). I see following changes required at the very least:

1. Replace all memory allocations in encoding/decoding with static pointer references which are allocated when the image info is first specified.
2. No need for Region-Of-Interests.
3. Don't add any EXIF or redudant header info to encoded data, as streaming server/client will already have an understanding of basic image attributes (i.e. size, color depth etc).
4. Keep support for multi-threading/core to ensure maximum performance.

I'd really appreciate if the developers would get in touch so we can plan this out. I'm willing to contribute my time to this. Thanks!

[Raphael Schweizer]

We have a related master's project ongoing (http://webapache.imvs.technik.fhnw.ch/~christoph.stamm/reports/P7_2011_MultiscaleVideoCompression.pdf, http://webapache.imvs.technik.fhnw.ch/~christoph.stamm/reports/P8_2011_MultiscaleVideoCompression.pdf) but no tangible release plans yet. You are welcome to contact Stefan directly.

- Raphael

[MKZ]

Hi Raphael, Thanks for replying! The paper isn't in English, so I'm not sure exactly what the author is saying, however, it does seem like that he/she is adapting it for video compression. Is that true?

I'm interested in using it for static content i.e. pictures that are streamed over internet (i.e. very low fps). Example would be sending 3-4 frames per sec of a computer's desktop to remote location where its decompressed and displayed, like vnc/remote desktop. So we can keep core algorithm the same, just move memory allocations during each encode/decode call to some other method etc, so that the algorithm is as efficient as possible.

Thanks!