Re: New transform for video

Thomas Richter wrote:

?? Come on, if you claim to be better, you need to provide evidence for
that. I'm not saying you're not. I'm only saying that you need to do
your homework.

I wrote "0.7 dB below H.265 KTA 2.2r1".

Then there's a lot missing, at least (besides tests), namely the motion
prediction and motion compensation - unless you have a novel design how
to handle motion without blocks.

The new transform is compatible with bocks. If it is done on a 4x4
array the synthesis functions are almost like the DCT.

Sorry, I don't understand. What do you do at the block boundaries? You
need to predict from somewhere. Again, if you stop the filter there (at
block boundaries) no matter what type of "extension" you pick, it will
impact the performance of the filter. It's a general problem of all such
filter types.

It's a general problem of all transforms at edges. Regions can be any
shape. A region can be a block, or a set of connected blocks, or the
whole frame.

Something like this: an inter frame can be divided into 2 regions: one
region is coded independently, the other region is inter predicted.
The region which is inter predicted is divided into subregions each
subregion correspoding to particular motion vectors. The S step can be
done on each region independently. The P step would be done without
crossing subregion boundaries.

So if in a frame the camera pans and all blocks have the same motion
vectors, then the whole frame can be coded as one region and there
would be no block artifacts.

Sorry, but if I follow your algorithm, it operates on rows and columns
separately, so it is a separable transform, i.e. its 2D operation is
simply a tensor product of two identical one-dimensional filters. Isn't
that correct? What do I miss?

The new transform can be described in 1 dimension or in 2 dimensions.
In the application the 1-dimension description is used because it is
simpler. In Libima it is implemented in 2 dimensions. See
transform_new in file transformic.c
.

Relevant Pages

  • Re: New transform for video
    ... I'm not saying you're not. ... to handle motion without blocks. ... Similarly, yes I can run the 9/7 wavelet filter on 4x4 blocks, but do I want to do that? ... Something like this: an inter frame can be divided into 2 regions: one ...
    (comp.compression)
  • Re: New transform for video
    ... Your patent statement mentions the SPIHT, ... Then there's a lot missing, at least, namely the motion prediction and motion compensation - unless you have a novel design how to handle motion without blocks. ... after one level of the S transform: ... Again, if you stop the filter there no matter what type of "extension" you pick, it will impact the performance of the filter. ...
    (comp.compression)
  • IRT: A New Theory of Relativity
    ... IRT is a New Theory of Motion. ... The laws of physics based on a clock second and light path ... The one-way speed of light in free space based on a defined absolute ... length of a measuring rod is maximum in the rest frame of the aether. ...
    (sci.physics)
  • IRT: A New Theory of Relativity
    ... IRT is a New Theory of Motion. ... The laws of physics based on a clock second and light path ... The one-way speed of light in free space based on a defined absolute ... length of a measuring rod is maximum in the rest frame of the aether. ...
    (sci.physics.relativity)
  • Re: Computer being developed modeled after human brain
    ... pixels in the next frame to compute its actions. ... Because you don't grasp that detecting patterns in frames is not the same ... the eyes move another, and the robot is riding a bike, on a moving train? ... Everything is is in motion, but yet, how does it associate the various ...
    (comp.ai.philosophy)