Physical position noise for CCD chip for more resolution

[daqq]

Hi guys,

I've been thinking: If you vibrated a CCD chip randomly by a small amount (say, 0.5 of its pixel) in both the X and Y direction, could this be used to add positional noise to the chip, then captured full frames randomly, shouldn't it be possible through some clever math to achieve some spacial equivalent of what you get with dithering for an ADC - essentially gain a higher spacial resolution?

If you just have a stationary CCD chip you can sample all you want, you will still be looking at the same points. But with some small vibration device (seeing as you have point sizes of several um these days, a MEMS vibration thingie on the PCB with the CCD chip should do it) you should be able to actually shift the point of view for each pixel, adding some dithering.

Thanks,

David

[lukier]

It is already there

https://en.wikipedia.org/wiki/Superresolution

And instead of vibrating the sensor you can just move the camera, run your favorite multi-view stereo algorithm to estimate the poses and do the same:

https://www.robots.ox.ac.uk/~vgg/publications/2003/Capel03/capel03.pdf

[Someone]

http://image-sensors-world.blogspot.com.au/2014/11/active-pixel-color-sampling-story-goes.html

Some maths as to why its not such a great idea.

[miguelvp]

As far as vibrating the CCD chip, some high end cameras do that but not for that purpose. As I recall they do it to shake the dust from the sensor if any got in there.

But that was many years ago, not sure if they still do that.

[Mechanical Menace]

I'd try two tiny linear actuators (some sort of piezo device maybe?) that can move the CCD by 0.5 of a pixel. You've 4 positions to take samples from so four times the resolution, no fancy maths because you know the exact spacial relation between pixels already*, and obviously only one quarter the frame rate if you need the extra resolution in both directions.



*So just a matter of putting each pixels data in the right spot of memory.


EDIT: Just thinking about it something like this could be brilliant for people who want very high resolutions but don't need short exposures. Astronomical imaging for example. Especially if you could do increments smaller than 0.5 of a pixel. Need higher resolution? Well have 4 positions for each axis, 8, 16... Bigger pixel sizes on the CCD could even be an advantage in this sort of application.

[daqq]

Thanks for the replies. As with all my ideas, it has been done 

[Halcyon]

As far as vibrating the CCD chip, some high end cameras do that but not for that purpose. As I recall they do it to shake the dust from the sensor if any got in there.

But that was many years ago, not sure if they still do that.

Not normally in video cameras but I can confirm my Canon digital SLR does it. It's presents a message on the screen when powering off and it's audible.

[encryptededdy]

FLIR actually does this in order to get higher resolution from some of their higher end thermal imagers. They call it "UltraMax": http://www.flir.com/uploadedFiles/Instruments/Products/T-Series/ultramax-technote.pdf

UltraMax uses the natural movement of the human body to capture an image set in which each image is slightly offset from the others. This results in a wealth of data that is much greater than any one image could provide. The data is combined to form an image that includes many more pixels of the target, resulting in a resolution greater than that of the original camera detector. The data is also used to create a clearer image, since pixel noise can be reduced through comparing similar points in multiple images.

FLIR UltraMax captures 16 thermal images in less than one second. These are stored on the camera as a single jpg file, and will appear as one image when viewed on the camera or in software. In the FLIR Tools environment, you can choose to enhance the image resolution. This is the UltraMax functionality. The enhanced image will have twice the original resolution and four times as many pixels. All pixels still include radiometric data, just as with normal FLIR thermal images.

As an example, the following 1280x960 image came from a 640x480 sensor;