smartphone cam with 41MP: Nokia 808

Started Feb 27, 2012 | Discussions thread
OP Bilgy_no1 Veteran Member • Posts: 4,880
Re: All cameras do this already.

OK here's how I see it.

I think this technology works best with random noise. A photosite's data is registered as either signal or noise, depending on a certain threshold. If you use smaller photosites to combine into one pixel, the noisy photosites are averaged out in rendering the pixel. So there is still one pixel, but it's rendered on the basis of the signal of 4 photosites and the noise of one.

This process happens before the pixels are being registered in the image file.

Noise Filtering does something else: it calculates the value of a noisy pixel on the basis of the signals of the neighbouring pixel. This happens after the registration of the pixels in the image file.

The same thing would be to take the full resolution output of the 41MP sensor and apply noise redution to it and then downscale it to the same output resolution. It would be interesting to see how the result of that process would compare to the oversampled output.

Louis_Dobson wrote:

The simplest for of NR is averaging the pixels (as per here). You trade res for (lack of) noise. In fact most NR tends to be more sophisticated...

David Clarke29 wrote:

Louis_Dobson wrote:

At least, they do if you do your own PP. Simply turn up NR and you are, in effect, pixel binning.

That is why the D800 is 36MP, and the fact that many people don't get this is why some ill-informed Nikon fans are complaining.

More MP is good. It always has been!

In the meantime, MFT remains the sweet spot. MFT is already far larger than it needs to be for IQ, but any smaller and you have no DoF control.

Bilgy_no1 wrote:

There's a strong feeling that m4/3, with its 'small' sensor, should keep the pixel count low. Less pixels mean deeper wells, and hence better sensitivity.

Enter the Nokia 808 ClearView, a smartphone announced today. It has a 41MP sensor, on a 1/1.2" sensor (almost the size of Nikon 1 sensor; most XZ-1 class sensors are 1/1.63", and 5x larger than regular smartphone chips). But it does a neat trick by binning pixels ('oversampling') to eliminate the noise. Another advantage is that you can use 'lossless digital zoom', both in stills and video.

There's a whitepaper explaining the technology here:

This kind of technology could benefit m4/3 as well. Not so much for the lossless digital zoom features, but for the noise performance and detail.

What do you think?

Proof in the eating etc. Let's see if we can find some samples...

Excuse my ignorance of all things in phography/physics but surely noise reduction and binning are two entirely different things. Binning is doubling up to increase the well size - this of course,all things being equal, will either reduce noise or increase DR depending on how they are recombined, however noise reduction is a case of smoothing (and thereby losing definition) of areas containg random voltages posing as information - at its simplest the good being overwritten along with the bad??

As I say, I'm just the average Joe, so probably my take on it is too simplistic.

Dave. (UK)

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow