Canon DSLR banding and noise reduction workflow technique.

A

astropixel

Guest
Interested in thoughts on this workflow - ideas, improvements.

In an attempt to get a little finesse and detail into my Canon 1000D images while avoiding saturated colours to cover up noise, I set about developing a post integration work flow, and this is what I came up with.

Instead of using Georgs' Canon Banding Reduction script on the single integrated RGB or histogram stretched image, I used the Channel Extraction tool, extracting the RGB channels and rotating them individually, applying Georgs' script separately to each channel, then re combining with Channel Combination to produce a new RGB, band free, noise reduced image.

One of the things I noticed, is that applying the banding reduction script to the integrated or stretched image caused the banding to move around slightly - it was impossible to eliminate banding completely (banding is severe on my camera despite the latest firmware update, but some of the noise is due to not dithering properly). Now all I have to contend with is slight streaking in the red channel, which is strangely absent in blue and green (the red channel is most affected by banding also).

This process will also work, but I think is less effective, on an untouched integrated image. Yet to try it on a stretched image. In any case noise reduction is significant. Nice flat images up to the application of LRGBCombination and further color stretches. Keeping application of tools to a minimum and careful application of lum masks was helpful at this stage.

Interested in getting feedback. It would be nice to use a process icon for this.

Cheers.

Rowland
 
Hi Rowland,

I also played with the idea to apply the Banding Reduction to separate channels or even the RAWs. In reality, I did not actually do it due to time constraints.

- Applying Banding Reduction to RAW should give some benefit for integration, since it reduces the overall noise before integration. But it is probably a lot of work.
- I am not sure why doing it to separated channels is better than doing it on the 3 channel image - the current implementation handles all 3 channels separately. The only thing you cannot control separately is parametrization of the process - all parameters are applied equally to all channels. Maybe that makes a difference for you since the red channel needs stronger parameters than G/B.

Georg
 
Hi Georg. The banding reduction script is excellent used this way on my images. I suspect that each camera is different. In-fact, banding was so bad that the image was almost unusable in under exposed areas. I was ready to give up until I thought to separate the channels. Still without the script there is no way to resolve the noise problem. Great job. Some of the streaking in the red channel was removed by the script also. Very happy.
 
Working backwards from the stretched images I was playing with, and where I ran into problems with banding moving around, I now note that with CBR script applied to an ABE CC integrated (linear) image, there is no need to separate the channels unless you want to work on the red channel independently, which was what I needed to do without affecting the other channels.

So I guess that, depending on severity, yes it is possible to isolate a channel and work on it independently of the others. Otherwise not necessary. But I do note that CBR is better applied after ABE and CC. I've tried several other methods and always ran into colour problems later during post processing.
 
one thing i ran into with integrated images is that on multi-night images, the rotation angle could be slightly different for each night. this means the banding noise from each night is superposed at slight angles to the other nights, and it becomes impossible to remove completely. i suppose this is an argument for doing the banding reduction on the RAW files. i think i ended up creating a stack for each night, then doing the banding reduction, and then aligning the stacks and integrating those.
 
That's a good point and except for banding rotation wouldn't ordinarily matter. The issues seem twofold, including heavy gradient removal, in another post, color calibration has been difficult to get right. The red channel is the offender with 99% of the artifacts. I'll try as you suggest.
 
Back
Top