Dana Diezemann published her presentation "High Dynamic Range Imaging, A short summary" at Image Sensors Europe held in London in March 2020. Few slides from the presentation:
I would like to give a little bit more explanation on my MAGIC log pixel from my research work at Institut National des Telecom (spin-off NIT): 1. There does be an integration at low light level since each readout is followed by a hard reset. The PN junction of the photodiode needs to be recharged, if light level is low, the charging process can not reach the steady state, consequently the response is similar to an integration pixel. 2. There is NO traditional image lag since each readout is followed by a hard reset. But there is a motion blur in the dark area of the scene due to the reason explained at 1).
The key invention of MAGIC pixel is that it proposes a hard reset to a "correct" dark reference giving the possibility to realize on-chip FPN noise compensation. In practice the silicon photodiode in this case gives negligible FPN with a highly linear SF. And this hard reset removes also the conventional Log pixel lag. But the illumination dependent motion blur remains still a problem and needs more investigation. There is still a large room of innovation on logarithmic pixel which can be a very useful pixel design for 3D vision chip in the near future.
I would like to give a little bit more explanation on my MAGIC log pixel from my research work at Institut National des Telecom (spin-off NIT):
ReplyDelete1. There does be an integration at low light level since each readout is followed by a hard reset. The PN junction of the photodiode needs to be recharged, if light level is low, the charging process can not reach the steady state, consequently the response is similar to an integration pixel.
2. There is NO traditional image lag since each readout is followed by a hard reset. But there is a motion blur in the dark area of the scene due to the reason explained at 1).
The key invention of MAGIC pixel is that it proposes a hard reset to a "correct" dark reference giving the possibility to realize on-chip FPN noise compensation. In practice the silicon photodiode in this case gives negligible FPN with a highly linear SF. And this hard reset removes also the conventional Log pixel lag. But the illumination dependent motion blur remains still a problem and needs more investigation.
There is still a large room of innovation on logarithmic pixel which can be a very useful pixel design for 3D vision chip in the near future.
-Yang NI
It didn't seem to cover Sony's DOL-HDR and Quad Bayer HDR.
ReplyDelete