Saturday, May 10, 2008

3um Pixel - Sweet Spot or Not?

It came to my attention that German Image Engineering launched 6MP web site telling there is no need in having more than 6MP in comact digital cameras, as image quality becomes worse at higher resolution. The basic claim is that 3um pixel is the optimal pixel size for compact digicams, thus the maximal resolution should be 6MP for 1/1.5" sensor.

One problem with this approach is that technology is progressing and what used to be true a few years ago, might not be true tomorrow. For example, pixel binning, if made correctly, can dynamically scale resolution and effective pixel size.

Talking about low-ISO pictures, a high full well capacity is important to get noise free pictures. Historically, smaller pixels had low full well. But there are few solutions in works to provide a dramatic improvement here. Also, various HDR modes should make small pixels more attractive, to some extent.

So, it might be true that the old generation 3um pixels are close to the sweet spot for compact digicams. I'm less sure that this statement remains true in a few years from now. Among other things, it depends on when compact cameras switch from CCD to CMOS sensors, as most of these solutions are inavailable in CCDs.


  1. Always a silly argument. Image quality, even if measured quantitatively, depends on the metrics and the weighting used.


  2. different anonymous:

    Actually not so silly an arguement. The basic premise is that lenses really only support a certain blur spot size (~ 6u) going smaller than 1/2 this really doesn't get you much better resolution. The additional arguement is that these pictures are taken under similar lighting conditions so some of the things to optimize for shot noise really can't be taken advantage of (like greater FW depth). I think they did yhemselves an injustice and over extended their statement, I can't buy into the other constraints about bandwidth etc. But IF they had said "for photographic sensors the sweet spot is ..." then they do have someting. For machine vision and other applications no.


  3. Still silly. For photographic quality, bigger pixels typically are better. But, if you start considering lens size, weight, cost, etc. then you can get a "U" curve and a sweet spot. Again, as I said, it depends on the figures of merit and the weight given to each. If you define all these things incl. the sensor technology and system limitations, then you find a sweet spot for your application and your target customer. Making any broad statements about sweet spots is, well, silly.


All comments are moderated to avoid spam.