From a "this is a traditional camera" standpoint, megapixel specs aren't completely meaningless, but they're certainly not the be-all, end-all.
However from a "what can I do with this thing as a sensor?" standpoint, more spatial resolution is in fact more power. Take a look at light field mapping [1] and the Lytro camera [2]. The technique uses an array of micro lenses placed over a traditional image sensor to capture many small low resolution images from many slightly different vantage points. Then a deconvolution algorithm is used as sort of a reverse ray tracing algorithm to actually map out the way light is bouncing around the scene. As a result you can generate many many different images of the scene in post processing, with whatever focus characteristics you want.
In this case the optics are relatively cheap compared to traditional camera optics, but the hardware to handle the deconvolution isn't.
Edit: To your original point however, in theory oversampling can be used to make up for crappy optics using techniques similar to light field mapping. Some mobile image sensors have a microlens array similar to that used in light field photography (OmniVision comes to mind). However in practice this requires that your crappy optics be consistent in how they're crappy, which presents other challenges.
I'm with you, though. When someone builds a cell phone camera with 120dB of dynamic range I'll get excited.
I know the details and understand pixel oversampling etc. In the real world they produce acceptable photos for the type of device but it is nothing special and definitely over-marketed.
I used an 808 for a week when they came out. Wishy washy colours, constant over exposure and distortion due to the pissy little lens. Not what they were promoting.
Whilst I compare this to a DSLR, a cheaper compact can do a better job of being a camera.
Personally I drag a DSLR around everywhere with me but that's because I started doing that sort of stuff before we had mobile phones and have developed an appreciation of quality prints.
> Wishy washy colours, constant over exposure and distortion due to the pissy little lens.
With the exception of maybe the distortion I'd chalk that up to crappy dynamic nonlinear exposure heuristics trying to boost dynamic range. For an example, take a look at the "HDR mode" in this guy's datasheet: http://www.aptina.com/products/image_sensors/mt9v034c12stc/
To be honest I think exposure was the software as you could reliably underexpose it and get the result you wanted.
Colours, probably dynamic range as you state. Looking at the curves for a pure white photo suggests range of blue was a little tight which is what made it look like out of date Kodak ektachrome on a good day. That may be software again bit I'm suspicious as the sensor on my D3100 doesn't exhibit that problem. Perhaps its better quality control.
Blue curves being off says they probably did a poor job on the transmittance of their blue filter material. There is also a trend toward oversaturation in cheap consumer cameras. Makes the consumer say "oh, it's so colorful!"
With cheap sensors the name of the game is always "get it close, and find a way to hack it so most people won't know."
There is no reason to compare image quality to DSLR unless you can keep DSLR in your pocket. If you want to compare against Nikon, compare against compact cameras.
This Nokia will big 1/1.2” sensor and it really removes any reason to have compact camera.
However from a "what can I do with this thing as a sensor?" standpoint, more spatial resolution is in fact more power. Take a look at light field mapping [1] and the Lytro camera [2]. The technique uses an array of micro lenses placed over a traditional image sensor to capture many small low resolution images from many slightly different vantage points. Then a deconvolution algorithm is used as sort of a reverse ray tracing algorithm to actually map out the way light is bouncing around the scene. As a result you can generate many many different images of the scene in post processing, with whatever focus characteristics you want.
In this case the optics are relatively cheap compared to traditional camera optics, but the hardware to handle the deconvolution isn't.
1: http://en.wikipedia.org/wiki/Light-field_photography 2: http://en.wikipedia.org/wiki/Lytro
Edit: To your original point however, in theory oversampling can be used to make up for crappy optics using techniques similar to light field mapping. Some mobile image sensors have a microlens array similar to that used in light field photography (OmniVision comes to mind). However in practice this requires that your crappy optics be consistent in how they're crappy, which presents other challenges.
I'm with you, though. When someone builds a cell phone camera with 120dB of dynamic range I'll get excited.