deband filter in xvid

[turbojet]

ffdshow's debanding filter seems to be very effective at removing banding, blocks and stabilizing xvid's output. But using it, gradfun2db or dfttest's dither function on the source isn't very effective.

Considering ffdshow is required to get this effect and all other decoders suffer. Has anyone considered adding a deband filter like ffdshow's to XviD that's used after the video is quantised but before it's written to the file?

[Didée]

Not possible. On decoding time, the video would be all-noise a few moments after each keyframe.

[LoRd_MuldeR]

How could we apply a debanding filter after quantization at all? :confused:

Quantization happens after the pixels have been transformed to the the frequency domain already. So after quantization we don't have pixel data, but (quantized) DCT coefficients.

Furthermore for P- and B-Frames only the residual (i.e. the difference between the predicted image and the original image) is subject to quantization...

[turbojet]

Does the banding in xvid occur before the quantization?
Using a smaller quantizer usually produces larger gradients on walls but decreases the banding on faces.

The only reason I mention it because it seems like it cures most of the the artifacts introduced by xvid with no noticeable drawbacks when used in moderation. Plus the source code is publicly available and if possible adding it to xvid wouldn't require much knowledge of the encoder (I think).