Sunday, February 6, 2011

Resolution of stitched images revisited

I was recently asked while shooting at Mesa Arch by a fellow photographer who saw me use my wooden panohead jig what using this contraption gained me. I was reminded of this going through some shots from yesterday. I went to Eldorado Canyon state park in the snow storm to shoot some images of the awesome rock formations there (I'll post images later). I compared a guide image with a stitched image and was struck by the amazing amount of extra detail.

Here is the full image to guide you. On the left the guide image, shot at 18 mm, f/11, 1/15s and ISO 200 using my 18-55 mm kit lens. This lens is amazingly sharp if used correctly. On the right, the assembled pano, rendered at 65 MP (Full rendered resolution is about 95 MP, but I rarely render the panos that large) from 9 shots at 48mm and same settings as above. This is a 12 MP Nikon D300.

Guide image. The screenshots below are slightly left of the center.

Of course at this size, there is no way to see any difference. It just illustrates how well this technique works. The guide image is almost identical to the stitched pano zoomed out this far. This is essential to me as it is otherwise very hard to compose shots. In this case, I wanted the lines of the awesome Fountain formation rocks to diagonally go to the right bottom corner. To compare resolution, I upscaled the single shot NEF file to the 65 MP resolution and show the comparison below at 1:1 (click on the image for full resolution - you might have to click again to get it at 1:1 on your screen).

Left: 12 MP image upscaled to 65 MP. Right: 65 MP stitched image of the same composition.

Even at the zoomed out size of the screenshot that I placed inline above, you can see the enormous difference in resolution. Note that if you look at this screenshot 1:1 by clicking on it, you are looking at the equivalent of a 65"x100" print (I am assuming your monitor is about 100 ppi - the reduced size image inline above is equivalent to a 27x42" print). I don't need to tell you how well the stitched image holds up even at that size. Because this is crazily large, I was curious how well this translates to smaller prints. So I downscaled the stitched image to the same resolution as the single 12 MP shot. Again, click to see a 1:1 image.

Left: single 12 MP image. Right: 65 MP stitched image downscaled to 12 MP.

Amazingly, even here you can see better detail in the stitched image. This is happening because of the Bayer mosaic used in the sensor of this camera and some lens softness. It cannot actually record as sharp of an image as it pretends it can, making a downscaled 65 MP image look signifcantly better than a native 12 MP one. Note that the 12 MP image was sharpened appropriately in Lightroom. At the 1:1 size, you are looking at a print of about 28x43" so this is still quite a large print. It is hard to show this on a screen, but when I make test prints from these images at smaller sizes than 28x43", say 16x24" I still see an obvious difference. Of course, to many people that might not be that important. It is also not unlikely that only a fellow photographer will notice it at smaller sizes and normal people would have to be pointed to the difference. To me, the print gains a large amount of presence and reality. This method can be applied to almost any landscape situation - even waterfalls! - to improve resolution.

1 comment:

  1. "This is happening because of the Bayer mosaic used in the sensor of this camera and some lens softness."

    yep, because pixels ain't just pixels. After the AntiAlias filters and the Bayer demosiac involving the creation of pixels which weren't samples its a miracle its as good as it is.

    So yep, stitching is great because it makes your sensor bigger, and bigger is better ;-)

    (for the same reason I love my 4x5)