Relationship Between Sensor Size And Megapixel?

HI John,

In IP cameras, is there any direct relation between size of the CMOS sensor Vs Megapixel. For example can I use 1/2" CMOS sensor to built 2MP IP camera? please suggest


There's a correlation between larger sensors and pixel count. In other words, the higher the resolution, typically the larger the sensor is.

For 2MP, I think most manufacturers use 1/3", 1/2.8", or 1/2.7". Presumably this depends what sizes the major chip makers (Omnivision, Micron Aptina, Sony, etc.) are producing (I am not an expert here).

As for 1/2" CMOS, I am sure you could but it's not common for 2MP. You rarely see 1/2" CMOS sensor used but if it is typically it is 5MP or higher.

Technically speaking, there's no reason a 1MP camera couldn't use a 1/2" sensor, or an 8MP camera use a 1/4" sensor. Heck, look at your typical smartphone camera these days - a Samsung Galaxy S3 uses a 1/3" 8MP sensor; I can't find the exact spec for my older HTC Desire HD's camera, but it's rated 8MP and I know it's a lot smaller than that (probably substantially smaller than 1/4", in fact).

Problem is, the more pixels you cram in a given area, the smaller those pixels become, and the less light they can collect, thereby affecting image quality to a degree, but ESPECIALLY affecting low-light performance. The classic analogy is to imagine buckets collecting rainwater: the larger the bucket, the more water it can collect; similarly, the larger the pixel, the more photons, and thus the better image.

So while the theoretical limits are pretty broad, the practical, functional limits are a lot tighter.

And as John says, it's also a matter of what's readily - and cheaply - available. If a lot of cameras are using 1/2.7" 2MP sensors, then the manufacturer of those chips will crank out a lot more of them, thus reducing per-unit cost, thus making them more attractive for other camera builders.

At the upper end of that are the likes of Avigilon's Pro-series cameras, which use full-frame 35mm Canon sensors (the same ones used in Canon's professional-grade DSLRs, I believe) for models from 8MP up to 29MP, and using Canon EF-mount SLR lenses. Of course, you also pay a LOT more for those.

Anyway, short answer to your question: no, there isn't necessarily a DIRECT relationship between sensor size and pixel count.

"Problem is, the more pixels you cram in a given area, the smaller those pixels become, and the less light they can collect, thereby affecting image quality to a degree, but ESPECIALLY affecting low-light performance."

I still contend this is over-rated. Derek is doing a studying cross referencing our low light performance tests with sensor size to see the strength of the relationship.

That said, even anecdotally the top performing super low light cameras use 'normal' size imagers of somewhere between 1/3" and 1/2.7". The 'secret sauce' is the image processing not the imager size.

Oh definitely, image processing is a big part of it as well... I mean, the best low-light camera I've used so far is still the CNB VCM-24VF, a measly 650TVL 1/3" analog camera, but still far and away above pretty much every other 1/3" analog camera... the main difference, as far as I can tell, being CNB's "Monalisa" processor.

My main point was, there IS a point of diminishing returns when you start to cram super-high resolutions onto super-small sensors.

The larger the chip, the lower the yields (percentage of end products without defects) and the higher the cost (goes up linearly with the AREA of the sensor). Therefore, sensor makers try to make them as small as possible while keeping picture quality acceptable.

I have seen a few surveillance cameras using 2/3" chips. They were fantastically expensive (thousands of dollars for a fixed box camera). I am not sure how much those chips cost, but it seemed clearly to drive up the overall price significantly. Also, lens cost would go up significantly too.

We can compare pixel sizes to the attached optics f-stop to get a ballpark estimate of whether there is a significant degradation based only on pixel density. This assumes decent lens quality; poor quality lenses will under-perform the f-stop and pixel-size based calculations.

"... there IS a point of diminishing returns when you start to cram super-high resolutions onto super-small sensors."

It's known as the Rayleigh diffraction limit. Basically, a real optical system can't focus a point source of light down to a perfect point. Instead, a point source is blurred out to a disk of some size, known as an Airy disk. In today's camera systems, as long as each pixel is larger than the size of the optical system's Airy disk, the Rayleigh limit doesn't meaningfully degrade camera performance.

Wikipedia does a very good job of explaining this effect, both presenting the math and at the same time providing a quick approximation. At f/8, making pixels smaller than about 4 um across would result in diffraction-limited visible image degradation.

Looking over these posts, I see one reference to a 1/4" 8 MP (3,224 x 2,448 pixel) sensor, which would lead to pixel sizes on the order of 2-3 um. I'm surprised that the math shows the Rayleigh diffraction limit would have already begun to come into play for that 1/4" 8 MP chip with an f/8 lens. However, if this hypothetical system were lensed with a decent quality f/4 lens, the Rayleigh effect would not be significant.

Beyond the Rayleigh diffraction limit, it seems likely that the size required by microelectronics in order to do "things" well will have a much more substantial effect. That's because the designers will (probably, if conscientious) match chips to appropriate optics to readily avoid the Rayleigh diffraction limit. If the much-referenced relationship between chip size and image quality is real, I'd imagine it would be because a big chip is better than a smaller chip because it has more space to pack in other "stuff" (I'm not a microelectronics guy but let's speculate) to improve things like dynamic range, blooming, and sensitivity/efficiency.

What I'm trying to say is, a well matched optics will land the same # of photons per pixel regardless of pixel size (up to the Rayleigh diffraction limit), but it MIGHT be possible (although I'm not ready to stipulate it) that larger pixels can do more with those photons than smaller pixels do.

There's reason to be cautious. Manufacturers of high end gear try to find things that justify their cost premium, and large format imagers are widespread in high end gear. Looking at the golden age of analog audio, a lot of "golden ears" found effects that were unmeasurable with state of the art equipment, but helped buyers justify stratospheric cost premiums. In that vein, I'd want to see robust proof before accepting that larger format imagers are inherently superior to smaller ones.

In any event, we can easily compare pixel sizes to the attached optics to get a ballpark estimate of whether there is a significant Rayleigh degradation, or not.

While not doubting this is true, I am not sure how this applies to video surveillance. For example, you cite, "I see one reference to a 1/4" 8 MP (3,224 x 2,448 pixel) sensor, which would lead to pixel sizes on the order of 2-3 um."

Sure, but I have never seen any IP camera with a 1/4" imager do over 1.3MP resolution. The 8MP 1/4" imagers are more a reflection of cell phones / smart phones where low cost and super small size are critical. Yes/no?

I think we agree on the general point (e.g., "I'd want to see robust proof before accepting that larger format imagers are inherently superior to smaller ones."), but I 'd like to clarify this for the discussion in general.

I think one could safely say that larger format imagers are GENERALLY superior to smaller ones as far as a starting point for image quality (weakest link theory, all else being equal, etc.)... the main question becomes whether that superiority is worth the inevitable added cost (and in some cases, for some uses, the physical size - one is not likely to build a covert smoke-detector cam out of a full-frame 35mm chip).

Looking over these posts, I see one reference to a 1/4" 8 MP (3,224 x 2,448 pixel) sensor, which would lead to pixel sizes on the order of 2-3 um. I'm surprised that the math shows...

Hey Horace what formula/constants did you use to come up with 2-3 um per pixel? According to the usually useless but sometimes saving "actual" sensor size it looks me to like it would measure 4.0mm diag, 3.2mm wide and 2.4mm height leading to 1um/Pixel. Yes/No? Though its impact on your other calculation is over my head, I'm a real scatterbrain when it comes to Rayleigh. ;)

I didn't know what a 1/4" sensor was: was it 1/4" at the longest edge or 1/4" at the shortest edge? It never occurred to me to use the diagonal, but your approach makes a lot more sense. I calculated them both, rounded to the nearest micron, and presented my ignorance as a range of values. Looking back over the results, I realize that I also must have assumed that the pixels are square.

longest edge: (1/4"x25.4mm/in)/3224=1.97um~2um

shortest edge: (1/4"x25.4mm/in)/2448=2.59um~3um

diagonal: (1/4"x25.4mm/in)/sqrt(3224^2+2448^2)=1.57um

Thanks for clarifying the 1/4" diagonal: that's good information.

As far as the Rayleigh limit, I can't remember most of that stuff, so I just Googled "Airy disc" on Wikipedia. It gave the formulas, but I didn't have to calculate anything because it showed that, using the average wavelength for visible light, focused through an f/8 lens (which it said was a typical value for imaging on a cloudy day), the smallest pinprick of light can only be focused down to a disc that is about 4um in diameter. Since the image cannot be focused any more finely than that, then under these conditions, sensors with pixels that are smaller than the finest detail that can be resolved with the lens cannot actually provide images that match the "native" resolution of the sensor.

Relevance: While we can say it and it may sound convincing, no one has made a compelling case that bigger focal planes are better, except in apparently irrelevant cases in which pixel size approaches the Rayleigh diffraction limit. Don't pay the premium for hype -- test it and find out the truth.

In the words of Mark Twain, "It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so."

Although I had no idea, most IPVM members were probably aware that the advertised size of an imaging sensor bears little relationship to its actual dimensions. At best, my calculations are irrelevant because they're based on incorrect "facts."

This excellent discussion what-is-the-size-in-mm-of-a-1-3-sensor provides more detail.