Friday, February 15, 2013

Progress report...

So how's the next release coming along?

The logic of the user interface - with all the new options in place - turned out to be quite complex, and the number of supported cameras has grown (for the next release also the Canon EOS 500D and the new Leica M, based on pre-production DNGs), and files from all the cameras need to be tested and put through all the new options.

So far so good.

The most problematic camera remains the Canon EOS 5D mark II. It's the odd one out, with its Bayer filter slightly different. And that difference keeps creeping up and causing problems in unexpected places.

The Canons are the most complicated anyway.

They differ per model in the treatment of the black level and obviously in sensor size. The 350D and 40D use separate lines to measure the black level (stored in the Delta tags), the other models don't, and they all make extensive use of very irritating variables like ActiveArea, CropOrigin, CropSize, MaskedArea... all kinds of offsets you constantly have to take into account.

Especially with the new 'resize' option, where the photo is basically cut in half on both sides, I had a hard time creating a DNG for the Canons with also the correct information where all these offsets were concerned. One wrong value and Lightroom wouldn't show the DNG.

I also ran into a snag with the noise reduction and the green filter.

Turns out the algorithm of the green filtering doesn't sit well with separately noise reduced results being mixed. Contrary to all the other filters, reducing noise in say the red result (for use in the green filter) and not reducing noise in the regular and blue result, can actually lead to more noise and quite ugly side effects. My basic conclusion was that, when mixed for green, the values within the separate results are too much intertwined and dependent on each other for the final - green filtered - end result. Changing values in one result do not necessarily lead to an overall reduction of noise. The two other - non-noise reduced - results can block the effect or even make it worse. This is also partly due to the algorithm - to get to the green filtered result - which is quite different compared to the red and blue filtering.

I need to study this a bit closer though. I wanted to do that anyway, cause the green filtering can also turn quite ugly if the white balance is pushed too much towards the yellow (slider to the right). Want to see if I can improve that or if I need to adapt the strength setting.

For now I solved the noise reduction and the green filter by locking the sliders. Noise is best reduced (and without any problems) in the green result, if all the results (regular, red and blue) are noise reduced equally before they're mixed. In the end it means that the noise reduction in DNGMonochrome isn't very beneficial for green filtered results. Lightroom can most likely deal with that one equally effective.

I hope to figure out a better way though, because in the gradient filtering Lightroom will be more difficult (although not impossible with their gradient noise reduction option in version 4).

Overall I probably need another week to finish up this release.

No comments:

Post a Comment