Megapixels - Are More Better?


Hall of Famer
As everyone reading this site knows, new cameras are coming out with higher and higher megapixel (MP) counts, corresponding to smaller and smaller pixels for any given sensor size. Conventional wisdom suggests that this MP race is bad for image quality, and it is common to read comments from camera enthuasiasts who wish companies would focus on making lower MP cameras better instead of pushing for higher counts. Even camera manufacturer representatives like Canon's Chuck Westfall suggest that the main drive for increasing MPs is that of market force rather than improved image quality. Camera reviewers including Phil Askey, Editor of DPReview, have frequently written of their disappointment over pixel stuffing. Michael Reichmann of the Luminous Landscape displayed the typical reaction to the MP hike when he wrote "My first thought when I read about the W300 was – Good grief – 13.6 Megapixels from a 1/1.7" sensor on a body the size of a tin of Altoids. Will the madness never end?"

Yet for every person longing for the days of 6MP compacts, there is someone else decrying the "megapixel myth". These latter individuals suggest that having more megapixels is in fact better for any given sensor size. They point out that today's high megapixel cameras make better images than the low megapixel cameras of old. In fact, Michael Reichmann really liked the image quality from that W300 with its teeny pixels. Of course as technology has moved forward, comparing recent high MP cameras to older low MP cameras isn't a great way to get insight into the effects of pixel size. The more relevant comparison is that of two cameras featuring the same sensor size and generation of technology but different megapixel counts. An example of this would be the 22MP Canon 1Ds Mk III and the 12MP Nikon D3 or the 6MP Fuji F31 and the 10MP Canon G7. Many folks, including most of the major review sites, have done such comparisons. However, the common error is to compare 100% pixel level crops of the two cameras. Looking at the pixel level isn't instructive as to how two cameras with widely disparate MP counts will compare at the whole image level. While the camera with smaller pixels will have lower detail relative to noise at the pixel level, in theory the two cameras ought to have similar noise performance at the whole image level.

There has been plenty of discussion regarding these issues, but never as much as in the past two weeks since DPReview decided to add pixel density to their product database. The title for their announcement, "When Moore is Less," makes it rather clear that they feel that more MP can be a bad thing. Seems as though a hundred or so threads have blossomed in response, largely dominated by the same handful of engineers, physicists, and mathematicians. Most of these individuals argue that in theory, for a given sensor area, the more pixels the better. Resolution is increased. Noise should not be significantly affected at the whole image level, and dynamic range losses at the pixel level should also be mitigated at the whole image level.

My understanding of these technical discussions is limited both by my knowledge base and the time I'm willing to spend piecing together scores of threads across multiple sites. However, one particular example posted by John Sheehy yesterday caught my attention and is the reason for this blog post. In his example, titled "The Joy of Pixel Density", John presented a comparison of images taken with the Canon 400D (5.7 micron pixels) and the Panasonic FZ50 (1.97 micron pixels). He shot an ISO 100 image with an actual (not effective) focal length of 22mm on each camera. The subject distance and f-stop were the same for both cameras, and he pushed the RAW files to ISO 13,500 equivalent in each case.

I won't reproduce John's crops here, but I will explain the images he displayed in that post. On the left panel, the inset represents a 100% crop from the 400D image, with the larger image in that panel representing an upsized version. On the right, the larger image is a 100% crop from the FZ50 image, and the inset is a downsized version. The important point here is that the area of the bit of sensor used to generate the 400D image on the left is the same as the area of the bit of sensor used to generate the FZ50 image on the right. The FZ50, with its much smaller pixel pitch, has far more pixels in this particular unit of area. The images speak for themselves, and John has concluded that "the [FZ50] sensor is better per unit of area than DSLR sensors, because it has a higher pixel density, which is just a fact, that some people seem to have problems dealing with." Snarkiness aside, he makes a good point. To be clear, the example is not meant to demonstrate that the FZ50 sensor is better than the 400D sensor. After all, a much smaller percentage of the 400D sensor was used to make this image. However, the example suggests that for a given size of sensor (area in this case), the higher megapixel sensor will give more detail and less objectionable noise (at the image level).

John refers to his example as "preliminary". If you read through the thread, Emil Martinec plays the devil's advocate. While I don't understand half of what folks like John and Emil say when they're arguing with one another, I can say that John's demonstration is not consistent with what most of us believe we are seeing in the prints and on screen at the whole image level when comparing two cameras of like technology with the same sensor size and different pixel pitch. Again, take for example the Fuji F100fd vs Fuji F31 or the Nikon D3 vs Canon 1Ds III (giving the old F31 some credit here by presuming it is of the same generation tech as F100fd). When people who have both cameras in such a comparison report on their actual experience with prints, they almost invariably say that the camera with the bigger pixels has better detail/noise in low light at high ISO at any given print size. That certainly has been my experience in doing those sorts of tests. I can't get a high ISO G7 or G9 print to look as good at 8x10" as my old F31 images did. I don't believe the explanation lies in amazing in-camera noise reduction from Fuji (otherwise their F100fd 6MP mode would be able to match the F31 at same output size, which it can't), nor do I believe that the answer lies in deficient G7/G9 image postprocessing. Furthermore, while I can see how downsizing images can improve dynamic range, it doesn't equalize small pixel camera image level DR with that of large pixel cameras in my practical experience.

I can only guess that the discrepancy between the theory put forth and what I see in real prints comes down to fill factor, along the lines of what Dave Martin discussed in that thread. For the sake of making practical decisions when purchasing gear, I trust my eyes. Having looked at a number of cameras with equal sensor size and different pixel size (comparing current tech to current tech), my observations from a practical standpoint when considering image output at a given size are as follows:
<li>In good light at low ISO, apparent detail/noise is better with smaller pixels.</li>
<li>In low light at high ISO, apparent detail/noise is better with large pixels, though less so at the image level than at the pixel level.</li>
<li>Dynamic range and quality of color/tones is generally better with large pixels, especially at higher ISO values. Again, this is less the case at the image level than at the pixel level, but it is a noticeable difference even in the former case.</li>
<div class="justify">
To conclude, while I enjoyed John Sheehy's excellent demonstration, I'd like to see a well done test comparing whole images from two cameras with the same sensor size and different pixel size. Barring such a test showing similar results to John's, I'm inclined to believe that more MPs isn't always better.</div>

Originally published on the old Serious Compacts blog. Older comments can be found here: Megapixels - Is More Better?