It seems photographers everywhere are jumping on the bandwagon urging camera makers to end the “megapixel war” and focus on other areas, such as “lower noise” and “more dynamic range” ... especially in the high end dSRL space.

Do we need a 30 or 35mp Canon or Nikon?  After all, don’t we have enough resolution?

Good question ... and after considerable pondering (well at least thinking about it a little) my answer is pretty much yes ... we need more. Before calling me crazy, let me explain where I’m coming from.

Admittedly I’m not the most technically savvy person around.  But current cameras deliver terrific dynamic range and amazing noise performance.  The above image was shot at 1600 ISO with virtually no noise issues using a Canon 5D Mark 2.

What current dSLR cameras do not deliver is absolutely pure resolution - so much resolution that the lens becomes the weak link (which it should be).  Sure some lenses can’t match current sensors, but most do just fine. And not that they aren’t terrific  (I love my 1DsMark3 and 5DMark2) but the very base technology of current dSLR designs cripple the resolution and though the end results are really good, there is room for improvement.

The basic problem is there is no relationship to the analog image projected onto the sensor and the sensor itself.  Thus to show any subtleties in detail requires a substantial amount of over sampling of the information.  Compare this to printing - think about how many dots it takes to print a decent image with an inkjet printer ... literally millions of dots per square inch.  How smooth or sharp would any detail be if we could only print a few thousand dots per square inch?  The same concept applies to capture.  There is no correlation to “detail” and pixels, and if we want to reproduce the detail perfectly we have to use a large number of pixels to record small parts of the detail and then put those together.  How many “pixels” would it take to really record the detail available on this cactus?  A lot more than I had.

Ok ... the next question is always “but do we need it?”.  Good question.  Just because you don’t need it doesn’t mean I don’t.  I like the ability to record maximum detail and even my PhaseOne P45+ disappoints me quite often.  I have hundreds of lovely images taken over the last 8 or 9 years most of which fall apart when I try and print them large.  Sure I can stand 30 feet away and they look OK, but I’m not buying all of that bull about only photographers looking at things up close.  As I’ve mentioned before if the image has something to offer up close ... some beautiful subtle texture or some interesting detail ... people will enjoy looking at it close.  There are so many beautiful images out there that are rezzed up to 400% to print a nice large print, and then we get upset because someone wants to see the texture in the flower petals (which isn’t there) and walks up closer.  HEY ... don’t they know they’re not supposed to look at it up close?

I’ll agree that not all photographers need resolution.  But lets’ be honest  ... this is about landscapes.  That’s what I love, that’s what I like doing, and that’s what I like writing about.

Currently dSLR’s are crippled in two ways.  First, like almost all digital sensors, they use a Bayer sensor. Each pixel doesn’t capture the actual color but captures only one component of the color ... red, green or blue.  To arrive at the actual color complicated algorithms use neighboring pixel values to determine each pixels actual color ... in other words very highly educated guesswork.  (Pretty amazing actually).  So what if each pixel of a 20+ megapixel sensor was actually made up of four “micro pixels” - one “mini” bayer array - and the pixel was determined by just this set of mini pixels.  This means each pixel could record exactly the color hitting it - no more demosaicing and figuring out what color each pixel really was.  Would this be better than an 80mp Bayer array?  Perhaps not, but I do know that occasionally I see artifacts resulting from the demosaicing (that’s the technical term for that highly educated guesswork of figuring out what color each pixel is using surrounding pixel colors, and artifacts are the result of not quite guessing correctly) This wouldn’t happen if each actual pixel represented the color it sees.  Another design option is a Foveon type sensor that could actually achieve high pixel density.  This sensor is really 3 sensors stacked, and each pixel captures all 3 colors.  A 20mp version of this sensor would be better than current 20mp+ dSLR sensors if it could equal it in all other aspects such as noise and ISO.

Another problem with dSLR’s is moire.  It occurs when patterns in fine detail closely correspond in spacing to the spacing of sensor sites. One common place where this occurs are fabrics, where the detail of the woven fibers are similar in size to the sensor sites.  Demosaicing the file exaggerates  moire as it gives it a rainbow appearance.  To counter moire dSLR’s are crippled again ... basically we have a soft focus filter in front of the sensor ...  the AA filter.   I’ve seen some claim certain cameras have AA filters that are “too strong”, but in reality the AA filter is engineered precisely based on the chip itself.  I don’t know if you can make a “weaker” one (I don’t believe so because as I said they are precisely engineered based on the sensor sites size and density) ... but I have  read in several places now that if the pixel density becomes tight enough you no longer need an AA filter.  Kodak actually thought they had achieved this when they introduced their DCS 14n camera which didn’t have an AA filter, but as it turns out 14mp isn’t enough ... not nearly enough.

We may be getting close now and indeed for landscape photographers we could probably eliminate the AA filter in the current 20+mp dSLR’s now available. Modified 5D’s with part of the AA filter removed performed very well, and at 30 to 35 mp on a full frame dSLR, the AA filter may no longer be needed ... at least for most types of photography.  This would yield a substantial increase in resolution.  Perhaps not the technical limit, but perhaps a practical limit.

When shooting landscapes, it’s pretty difficult to know before hand how good the image is going to be let alone how large I may want to make it, but I do have a goal of achieving a great image which also contains amazing detail and looks fantastic when printed very large. As I mentioned, dozens of my favorite images fall apart if I print any larger than 16x20 (which to me is quite small), and some don’t even look good at that size.  Things are much better now with current cameras, but there is still room to improve.

OK, now you can call me crazy.  :)