The cube-shaped module is about 1 centimeter per side and contains a dense array of 500,000 lenses, each 30um in diameter, in front of an image sensor measuring 5 mm by 7 mm. Each lens captures a slightly different image from one another, and the camera is said to produce a large, complete picture by using original software to combine the 500,000 tiny images.
The new camera is said to be able to measure the distance to an object based on the differences among the small images. It can also create pictures that are focused on every single part of the image. The camera module can also be used to take videos, and allow the users to retain the image of a figure in the foreground while replacing the background.
Toshiba plans to commercialize the module by the end of fiscal 2013 (March 2014), and call on smartphone manufacturers and other companies to adopt the technology.
Toshiba provided the newspaper with the sample images showing the camera module's after-the-fact refocusing capabilities:
 |
| A picture refocused on a man on front |
 |
| A picture refocused on a man on far back |
Credit to Toshiba for sticking out their neck. They were also the first major company to embrace and market CMOS active pixel sensors with PPDs.
ReplyDeleteOn the other hand, the images really aren't too good at this time and I am not sure what the value proposition is. Time will tell I guess.
I am not quite sure what is the advantage of this technology over Extended Depth of Field (EDoF), (e.g. similar to what is proposed by DXO and other vendors).
DeleteClearly, light-field imaging has an advantage in generating 3D images and could also be useful for surveillance imaging. Yet, I don't see the merit in integrating this technology in the mass CE market: the computational requirements are huge and will grow exponentially with the increase in resolution, (which is currently rather poor). Seems as a gimmick at this point (much like 3D imaging using dual-sensors). What will determine the market adoption is the cost of these sensors. To me it seems that the manufacturing costs of such micro-lenses arrays is very high - much higher than producing the micro-lenses that are placed over pixels in CMOS imagers.
The cost vs. the effectiveness seems very high but, as Eric wrote - time will tell...
Eric, you are right, but in the early '80s the quality of the CCD images was not that good either, and in the mid '90s the quality of the CMOS images was not that much better. For sure the quality will become better over time.
DeleteNot the most impressive image quality.
ReplyDeleteI wonder what the max processed image resolution is. I also wonder how the processing algorithms will be supported on multiple processors.
Also wonder if they process images in realtime for preview and video?
The newspaper writes "by using original software to combine the 500,000 tiny images." So I guess their prototype processing is not real time yet.
Delete1 CM thick for mobile or tablet ? That is strange. modules are 5-6 mm today and OEMs are pushing this down.
ReplyDelete10 mm thick for mobile or tablets sound irrelevant.
ReplyDeleteTest images look very poor and certainly not acceptable quality for a mobile phone
ReplyDeleteLightfield technique means that a primary lens is required in addition to an MLA (Micro Lens Array)
Means modules are still thick resulting in a bump on the back of the phone
Low-light problem isn't solved either
Computationally very expensive (Favaro http://goo.gl/B2ZhB) unless you are prepared to accept loss in resolution (Giorgiev)
May not be an issue if you use a 41Mpixel PureView sensor as a basis but conventional sensor technology means cross-talk issue still doesn't go away
Generating stereo views is bound to disappoint (https://www.lytro.com/camera) owing to the very limited parallax (width of sensor)