Can the macro block orientation bug be fixed in software?
 

Hi,

I am going stir crazy waiting for the new AVCHD cameras to come out, or at least a quality low cost 720p 25fps camera. But it got me thinking, is the macro block orientation problem with the HD1 that bad, can it be rectified by software processing? I don't know anything about how these compression systems and their file formats work, but the macro block references should be stored discretely somewhere it can be changed the right way around.

Here is a link to the bug that was just posted a few days back by Carlos Serrano:

http://www.i-real3d.com/Archivos/HD1Bug.jpg

In:
http://www.dvinfo.net/conf/showpost....38&postcount=8

So, is it possible to change the macro blocks in the recorded files around to correct the bug?


Questions:
How much does it really effect compression? The orientation might even not affect the compression of the image at all, it might be separate from compression. One thing is for sure, it should cause trans-coding quality to drop.

Would correcting it produce a big benefit to image quality? I wonder how much of the problems in details and waves is the result of this bug, and how much it could be corrected with a fix.


Thanks

Wayne.

Looking at that picture, is this an orientation problem, as previously suggested?

From what I can see, I think what may be going wrong is some kind of absolute value style error in the angle. If this is the case, then you can either have the angles from 0-180 degrees or 180-360 degrees, but without some access to the compressor chip, you wouldn't be able to tell what to switch to the new side.

I'm not certain though. I am really hoping for Sanyo to just release an updated firmware.

From the reply in the other thread, this might turn out to be the only solution. If it is evident in the file it is recorded wrong then it would be very simple. If it is not, simple software analysis should be able to tell that an object is bent is a certain direction, and be able to clean up most of the blocks around that object, with more detailed analysis of details (making the detail from one block run into the matching detail of the next block, it should be bale to restore most of it).


Here is an interesting idea, you might not be aware that noise and other differences in an image produce differences in between frames, with different details being revealed in different frames. Processing the footage to look over frames to gather missing details and make a better quality recoding of the original, you might make a recording "at least" the quality of DVCPROHD, or XDCAM HD footage, from a $799 camera. Interesting isn't it.