A Case for More Pixels?
After my earlier post about the 0.22 megapixel Nikon D800e, it might seem inconsistent to make a case for more pixels, but I just stumbled on something worth sharing — a noise reducing strategy I'd not seen before. BTW, I think I'm very late to this party, but better late than ignorant forever.
I was having dinner with a friend, Dave Best, and showing him my newly arrived Panasonic GM1 — the micro-sized travel companion for the new larger camera. We were snapping away, just playing and goofing off as you do with a new toy when it arrives. I had set the camera to ISO 3200 and JPG just for grins and snapped a few shots of David sitting across the table from me. Next day, after loading the images into Lightroom, I was flabbergasted at the quality of the JPGs. Simply stunning — the image quality, not the image content nor the model (sorry, David). Here is the full image and a few 100% screen grabs of Dave's beard, jacket, and the out of focus ceiling that would normally be a place I'd expect lots of noise. Wow.
Keep in mind, this is ISO 3200. How could it possibly be this good? Filled with kneejerk doubts, I double check the EXIF data, and sure enough, ISO 3200.
I then noticed that in my haste to set the camera controls — and not being fully up to speed with this newly arrived camera — I had inadvertently set the camera to record half the sensor's full potential. That is, it captured 8 megapixels rather than the sensor's native 16 megapixels. In essence, the image is downsampled in the camera. Could it be that the in-camera downsampling did something to limit the expected high ISO noise?
A quick bit of research on the Internet and sure enough I found that downsampling to reduce noise is a hot topic — most of which went right over my head in geekspeak and science stuff. I'm a photographer, not an engineer, so I'm not in the least qualified to analyze or comment on the validity, theories, debates, or the hows and why's. All I know is that the photo of Dave was a mind boggling, low-noise, sharply detailed success at an ISO I would have predicted complete and fuzzy failure. I've always thought of noise reduction as "detail-squelching" and the finished result a compromise of competing wishes. Here was an image that seemed to offer the full muffin and the ability to eat it, too. This was worth investigating.
I read about as much Internet contentiousness as I could stand with competing opinions and science/pseudoscience about all of this, and came away more confused than I started. It seemed simple enough: In theory, if downsampling is a strategy to create images with less noise (for whatever scientific and engineering reason), it gives us photographers another tool in our arsenal for that difficult situation where we are needing all the help we can get. The exigent tools are a fast lens, image stabilization, and large sensors. Maybe now we could add to that downsampling as a part of the low-light strategy we need for success. Testing is in order.
I popped off a couple of exposures in our dark parking garage, both shot at ISO 6400 — a JPG and a RAW file. The JPG was downsampled in the camera to 8 megapixels.
Below is a 100% detail crop of the JPG . . . again, wow. I don't care why or how, just WOW! (I can hear some of you who've known about this for ages snickering at your monitors as you read this. As I say, I'm late to the party, but at least I'm here with my dancing shoes on.)
ISO 6400. Handheld at 1/15th second at f/8. What noise? ISO 6400, for goodness sake! Amazing.
For me, the final test is always in the print. Both images print beautifully at the modest sizes I typically use, 8x10 or thereabouts. I even printed them at 11x14 and both were pretty good — not perfect, not stunning, but solidly good. I don't do giant prints as a rule, so caveat emptor at larger sizes. At my sizes — 10" or so — very useable results even for critical images.
Which brings me back to the Nikon D8003 and its 36 megapixels. Theoretically, if you only need to make an 11x14 image or so, you will need 4200 pixels on the long side which is about half its native sensor resolution. With its full frame sensor to begin with , does this mean in-camera downsampling would yield stunningly printable images at ISO 12,800 or higher? What a mind-warping concept for a photograph who grew up thinking ISO 1600 meant Rodinal overnight and Tri-X grain the size of golf balls.
Finally, what about shooting RAW rather than in-camera downsampling of JPGs? Here is the same subject, photographed handheld at the same time (also 1/15th sec at f/8) using RAW and ISO 6400, opened in Photoshop using Camera Raw and downsampled there to 8mp, doing visual noise reduction and sharpening the way I would normally do so. Here's a 100% crop. Looks pretty good to me.
Just in case you're wondering, below is a 100% screen grab of the naked RAW file before any processing . . .
And finally, below is the RAW file from the above after a little Noiseware Pro noise reduction, still at 16mp, full sensor capture.
I'm not at all confident that I understand this downsampling business from a technical point of view, but I like it. Wish I had stumbled onto this a long time ago. Probably should read up on it and do some more testing. All I know at this point is that I would not hesitate today to use ISO 3200 and maybe even 6400 with in-camera JPG downsampling if I needed to, or I could shoot RAW and simply downsample in Camera Raw as I open the image in Photoshop. I'm yet to see if I can accomplish the same thing in Lightroom. I'm also yet to use this technique in a real-world project, but I'm looking forward to trying it.
Any of you have any experience with this concept you'd care to share?
Brooks' books on photography and the creative process are available in print from Lulu.com, and as eBooks for Kindle or EPUB readers. As one of the membership benefits, these eBooks are available in their entirety to members of LensWork Online via download.