Why does higher resolution mean that you tend to get more blurry images?

I’ve generally just taken Thom Hogan’s word for it that higher pixel density requires more shot discipline in order to avoid blur. After all, it’s sort of intuitive that slight movements will be captured by finer grained pixels more easily than by coarser grained pixels. But the D7000’s woes on its forum at DPReview indicate that not everyone is a believer.

So I thought that it might be useful to explore the situation and find a plausible layman’s explanation for it. Bearing in mind that I am a layman and thus this is my way of explaining it to myself.

So let’s start with a single pixel.

Yes, that’s one big pixel. Pretend that it comes from the 4mp D2Hs, where the pixels were almost that big :-)

Now let’s add a detail that is going to be captured.

This detail will actually cause this pixel to change color to a blend of the two with proportionate contributions of the color. Or something like that. The demosaic algorithm will take care of it by interpolating the color from this pixel’s color and that of its neighbors.

Remember that each pixel is actually monochrome luminance that is filtered through one color in the Bayer mask for most sensors. That color with the neighboring colors allows the demosaic algorithm to make excellent guesses at the color of this pixel. And that algorithm repeats for every pixel, which is why we get a pretty decent level of color detail.

This fine detail is not really captured of course by the sensor since it is smaller than one pixel (one unit of quantization), so the image will not appear to perfectly match the scene at extreme magnification. But you have to go to pretty low resolution for that to matter much in a practical sense.

Let’s push on.

Now let’s move that pixel as the result of hand shake during capture.

No big deal. The detail is, after all, still within the bounds of the one pixel. Which means that there is not really enough movement to see a detail blurred.

Now let’s take this pixel and divide it into four pixels, which in fact is exactly what happens when you go from a D2Hs to a D7000. From 4mp to 16mp.

Hmmm … this little detail was mostly contained in the top left pixel before the blur. But when there is a bit of shake, it starts to make a more substantial contribution to all the pixels, and especially to the bottom right pixel, which will be 50% the wrong color now. That is probably going to be noticeable.

Same detail, same blur. Fairly different result.

Now here is the thing … if you were to take each image and downsize to 800px for Facebook, the blur will likely vanish. But if you were to take each pixel and upsize for a poster, each image responds differently.

• The blur does not appear at all in the 4mp image, because individually coarse but sharp pixels are simply interpolated to fill the space on paper. It looks a little soft but from a distance it looks very sharp.
• The 16mp image, on the other hand, looks outright blurry from any distance. The pixels are mushy at the borders and to go onto a poster, this gets magnified somewhat to fill the space. A large gallery image might be 40 inches on a side, which would require that the 4800-odd pixels on the long axis be almost tripled to 12000 for 300ppi. That’s a fair bit of magnification of all that blur.

So it turns out that shot discipline really does matter. And note that we’re talking tiny amounts of blur here. This is not the big blur that shows up on all your images when you try to shoot without VR inside a dark room. All cameras see that.

This is that microscopic blur that was never visible on your 4mp or 6mp dSLR but is always showing up on your 16mp or 18mp dSLR, driving you completely nuts. The kind of blur that you get from being lazy. Not bracing when you shoot hand held. Not using a tripod with a cable release or timer. Not using mirror up when shutter speeds fall in the danger zone. That kind of thing.

Just watch the forums for a while and you will see what this does to people until they realize that they need faster shutter speeds than before. Exactly twice the speed in this case, because the blur tolerance is exactly half the previous tolerance on both the X and the Y axis (i.e. horizontal and vertical shake.)

I hope that helped a bit …

Actually, I hope that’s fairly close to right