Thanks for your contribution Nick. I'll wait another day or so to see if anybody else wants to chime in before revealing the order.
Fair enough. I'm actually quite curious about the demosaicing algorithm used on the center image. It's extraordinarily artifact free for a single frame. The left is very much in line with what I'd expect from a single frame, and the right is the pixel shift, but the center is a bit of a mystery to me.
P.S. How do you perform a blind deconvolution/demosaicing process on a single source image?
Right, there was no sharpening or other processing of any kind. By blind I meant that one raw conversion made no assumptions whatsoever, while one was neutral target prior aware.
I understand now.
I guess it's what we could expect in the future by really smart algorithms.
Yes, it will be almost as good as what we can expect today from well-implemented pixel shift.
Good point. And without further ado, given the great interest demonstrated for the K-1 in this thread ;-) , congratulations Nick! Your analysis was spot on (applause), my observations were very similar.
The demosaiced file, first to the left, was rendered by RawTherapee with the dcb algorithm and its enhancements and CA enabled. It typically does a great job without photographer intervention, as it did in this case. I think I was not able to defeat the initial color matrix adjustment in RT (though I was able to kill profile and tone curves), which desaturates blues in the current strategy of DNG v1.4, that's why the blue channel is, err, f'd.
The image to the right is produced straight from the fully populated DNG converted 'LinearRaw' channels with no processing whatsoever other than white balancing and application of sRGB gamma. Impressive.
The grayscale image in the center is simply the unshifted cfa raw data as-is, except for channel normalization (white balance) and application of sRGB gamma. I am glad it gave you (and at least another nameless poster in the other thread) pause and you'd consider it extraordinarily artifact free for a single frame. In a double blind test I would expect it to be virtually statistically indistinguishable from the pixel-shifted 4 captures when viewed by the unalerted typical photographer pixel peeping at 100% - unless clearly put under a forensic microscope by someone skilled in the art, like you. Of course this approach only works in the neutral portions of an image.
If someone can provide a separate single frame from the 4-frame pixel shifted sequence of IMGP0551.PEF in the next couple of days I will pit the two versions against each other using the new RawDigger beta and DNG converter.
In the meantime I think I have now been able to touch first hand the advantages of pixel shift technology. It seems to work, good job Pentax. As to whether I would use it on my tripod mounted landscape captures given the massive files and additional 3 second in-camera rendering time, I am not sure yet. I would probably want to compare these results to 4 unshifted captures assembled by something like
photoacute after a bit of subjective processing before making up my mind.
And good job Nick!
Jack
