mike's blog

contrast, contrast, contrast....

Hi folks,

good news; the next filter is on its way... and what should I say, it's again a local contrast filer... up to know, I couldn't find the magical local contrast filter, which suits in all situations, so gradually step by step... this one amplifies textures, so contrast in the small structures... like always a sample, which shows before and after that filter, so this is the only difference... enjoy Wink

greets mike

dlRaw: actual denoising status

Hi folks,

the people from digiKam have implemented the actual version of wavelet denoise (the same as dlRaw Wink) and with this they posted some comparison images, you can see them here 1, 2, 3 and 4. The RAW file is freely available (thanks to the guys from focus-numerique) so I gave it a try on my development version...

greets mike

Image quality

Hi folks,

just a few lines to let you know, that the project is not dead Wink So in the last days I've worked on one important part of RAW processing, the demosaicing. Additional filters were added, and with these image quality can be improved.

Demosaicing is the part which creates from the Bayer pattern the actual image right at the beginning of the processing. This is important when it comes down to details, noise and sharpness (and of course artifacts Wink). We're mostly talking on pixel level here, so the differences are not that big (100% view in the 1:1 pipe). I've prepared two small losslessly compressed samples (the RAWs are kindly provided by imaging-resource, I took the ISO 100 and ISO 25600 shots). So just download the files yourself and play with the settings on the 1:1 pipe (otherwise the demosaicing filters won't affect the output) to see if you can get similar results with the current version without these algorithms Wink Only settings on the camera tab are used, no additional sharpening or denoising, and no wavelet denoising on the RAW. Sure it is not perfect, but an improvement. You won't see the detail differences very often, because you really need fine structures for this, eg small patterns (moire areas) of fine bright hairs (like whiskers)... The differences in the noise on high ISO shots are far more prominant. Of course improved quality doesn't come for free, the calculation times are much higher, but it is still bearable with a modern multicore CPU, all of them are already parallelized (demosaicing of 10 MP on a Q6600 with very good quality and many refinement steps takes around 10 seconds). For more information, there is a nice comparison of demosaicing algorithms here. So, have fun to explore your RAW files for these fine structures and stay tuned for the next release Wink

greets mike

dlRaw: hotpixel reduction on bayer pattern

Hi folks,

just to let you know, the first filter for the next release is working. The automatic hotpixel reduction works now directly on the bayer pattern, to kill bad signals before they get amplified by the demosaicing algorithm. At the moment it works quite conservative to preserve as much detail as possible, but this comes with the cost, that not all are catched... but judge yourself, a small 100% sample of a night shot with my K100, demosaiced with AHD and sharpened with unsharp mask (radius 2, amount 1, threshold 0 ). The difference between the two is just the hotpixel reduction. Stay tuned Wink

greets mike

 

dlRaw: hotpixel reduction

Hi folks,

just a short update, to let you know, that work is still preceding... You probably noticed my last pictures on the blog. They are pretty clean as you see them there, but while processing them I had a problem with hotpixels. My old and trusty Pentax K100D has some bad pixels when used with higher ISO. Although it's not so bad on the pure picture, when sharpening the image bad pixels like to shine out Sad So I had to remove these small blemishes on all of them manually in gimp afterwards... Of course I didn't like that Wink During the week I tried to code an algorithm removing them; finally today I can report first results.

The algorithm works at the moment only on 1:2 and 1:4 images, because on 1:1 pipe it is better to work on the Bayer pattern directly and that is still work in progress... But on 1:2 I'm pretty satisfied with the result. The two attached images are both way oversharpened, that is just to show you how many bad pixels there are. Usually, with normal sharpening, one would only see a view of them, but those would still be distracting. The difference between the two pictures is just the hotpixel reduction nothing more.

So stay tuned for more news about dlRaw Wink

greets mike