Tuesday, May 12, 2009

Why do I want a low resolution camera?

These days it's not uncommon to hear about consumer cameras with 10+ megapixel resolution. To the average consumer this seems like a good thing. How can more pixels be anything but good, right? In reality there are several tradeoffs to using more pixels. This is especially true for any scientific imaging (this really applies to low light an/or quantitative imaging such as microscopy and astronomy). So what are the downsides?

First let's look at the concept of a pixel. Pixels are physically sized on a CCD or CMOS sensor. For an example of pixel architecture check out this page. The physical size of a pixel affects several aspects of image acquisition:

  1. Dynamic Range
  2. Sensitivity
  3. Resolution
  4. Speed
Most people simply think "bigger pixels means less resolution" or "more pixels result in more resolution". This isn't the only thing that is affected by pixel size. A more interesting question to ask would be "What am I sacrificing by using smaller pixels?". Of the four performance aspects above three are negatively impacted when using smaller pixels. These are:

  1. Dynamic Range
  2. Sensitivity
  3. Speed
Dynamic Range

The dynamic range of a sensor is first determined by the full well capacity(FWC) of a single pixel. This is a measurement of how much energy the pixel can hold before either becoming nonlinear or before the energy spills out of the pixel into neighboring pixels (called blooming). Smaller pixels simply cannot hold more energy due to physical size, so smaller pixels = less dynamic range. Let's compare two common chips - the Sony ICX-205AL CCD, used in the Photometrics CoolSNAP cf2 camera (and many others), with the Sony ICX-285 CCD, used in the Photometrics CoolSNAP HQ2 (this ccd is the most prevalent in modern microscopy cameras).

  • The 205 CCD shows a FWC of 9,700 electrons.
  • The 285 CCD shows a FWC of 16,000 electrons.
  • Note that the 205 chip has a pixel size of 4.65um, while the 285 has a FWC of 6.45um.
In the case of these two sensors, a 2x increase in pixel area provided a 65% increase in FWC. The larger chip has more FWC, so I can detect a range of brightness with more accuracy, or have more range of bright to dark, or more dynamic range.

Sensitivity

This is an easy one: If I take a given area of my image being projected onto my chip, I chop that area (or that brightness value) up more times if I have smaller pixels. Lets assume I have one chip using 5x5um pixels, with another that uses 10x10um pixels. Let's also assume the light being projected onto the sensor area is 100 photons per 100um square area. Using my smaller pixel chip each pixel could collect 25 possible photons. Using my large pixel chip each pixel can collect 100 possible photons, so the larger pixel camera can grab more light per pixel than the small pixel camera.

Speed

The argument I make here isn't really a technolocal limitation but more a common product/market limitation. All of the cameras we use are set up to collect an image projected from a microscope into roughly a 1" diameter area. So I am limited in how much area I can use on my chip. Because of this if I have bigger pixels I cannot simply make a gigantic chip that has say 1000x1000pixels, as it's possible the corners of my chip won't have any light projected from my scope. This means that intrinsically I'll either find chips with a lot of small pixels, or chips with fewer large pixels.

When considering speed there are several factors: exposure time, analog to digital conversion time, and "frame shift" time. (I'm using this term to conver interline shift and FT operations for those purists out there). Now if you look at any data sheet you'll notice there is some "top speed" a camera can go. You can bin the camera and use a small readout array but you hit a limit - why?

Let's assume we have a camera of 100x100 pixels that can expose an image and get enough light in 10ms. Let's also assume our A/D converter that turns electron values into digital numbers can operate at very high speeds, say 100Mhz. In a case like this our camera can theoretically run at 100 frames per second, which is quite fast. Now what aspect of the system will stop me from say driving the camera at 10,000 fps? If we consider that the camera must move each pixel's energy off of the chip, using shifts of energy from one pixel to another or into some series of registers, whatever time it takes to move from pixel to pixel will become important. This is referred to as the shift time or shift rate. Depending on the architecture of the camera this time will affect the camera to different degrees, but ultimately this is a measurement of "How fast can I move energy across my chip?". This bucket-brigade type of energy transfer is what takes a electron charge in pixel 1A, and spits it out the other end of the chip into the A/D converter.

So using our fake camera here let's say it takes our camera 50 microseconds to move from one pixel to another. Then let's consider how many moves must be made to get pixel 1,1 out the other end of the chip at position 100x100. Basically pixel 1,1 needs to be moved 100 pixels down and 100 pixels over. So 100x100=10,000. Each shift = 50us, so 50us*10,000 = 500,000us, or 500ms. So using our fancy super fast camera we can only go 2 frames per second!!!

Obviously the example above is using excessively show shift rates so I can make my point. The bottom line here is that shift times do have an impact. So how does this relate to big pixels? Well if I have bigger pixels we already found that I must use less of them. Less pixels = less shifting that needs to occur, so the impact of shift rates becomes less! Ultimately you'll find that cameras like a 128x128 EM system can go up to 4000 fps, whereas a 512x512 EM system only goes up to 500fps, with the difference being shift rate*pixel count.

Conclusions

So many times I am asked why big pixels could possibly be better. Now you know! Don't let the hype of "megapixels are cool" lead your lab into an inappropriate camera for your research. Remember that the camera is a detector, just like other instruments. While the appearance of the image may suffer with large pixels, the data contained in the image can be more informative and accurate than data collected using high resolution systems. Try to prioritize what performance aspects will aid your area of study, and keep these aspects in mind when finding the right camera for your work. You'll be happier in the end with the right tool for the job.

- Austin

No comments: