Thursday, November 12, 2020

Limits of HDR Imaging with Quanta Image Sensors paper "HDR Imaging with Quanta Image Sensors: Theoretical Limits and Optimal Reconstruction" by Abhiram Gnanasambandam and Stanley H. Chan from Purdue University shows DR advantage of QIS over CIS:

"High dynamic range (HDR) imaging is one of the biggest achievements in modern photography. Traditional solutions to HDR imaging are designed for and applied to CMOS image sensors (CIS). However, the mainstream one-micron CIS cameras today generally have a high read noise and low frame-rate. These, in turn, limit the acquisition speed and quality, making the cameras slow in the HDR mode. In this paper, we propose a new computational photography technique for HDR imaging. Recognizing the limitations of CIS, we use the Quanta Image Sensor (QIS) to trade the spatial-temporal resolution with bit-depth. QIS is a single-photon image sensor that has comparable pixel pitch to CIS but substantially lower dark current and read noise. We provide a complete theoretical characterization of the sensor in the context of HDR imaging, by proving the fundamental limits in the dynamic range that QIS can offer and the trade-offs with noise and speed. In addition, we derive an optimal reconstruction algorithm for single-bit and multi-bit QIS. Our algorithm is theoretically optimal for all linear reconstruction schemes based on exposure bracketing. Experimental results confirm the validity of the theory and algorithm, based on synthetic and real QIS data."


  1. Congratulations Abhiram and Stanley on laying a strong mathematical foundation for HDR image reconstruction using the QIS concept and showing both theoretical and experimental results. What is also cool is that this works for SPAD-QIS and CIS-QIS implementations (at least for 1bQIS).

    1. Even though I am involved in SPAD-QIS, my interest in CIS-QIS is growing by the day.
      Professor Fossum, if I may, I have been going through various of your publications lately to get a better grasp on the CIS-QIS physics and challenges. Although I cannot claim to have yet an in depth understanding of it, my attention was drawn lately on some work you did on the readout part involving a single electron transistor.
      To my understanding, this work was part of a collaboration with Samsung. Is there some accessible material out there related to that work or has it not been disclosed for the obvious reasons?

    2. @Yannick - Thank you for your interest.
      I agreed to join Samsung as a consultant (albeit at about 30% FTE) in 2008 to be able to work on the QIS challenges although much of my effort was focused towards iTOF with a goal of RGBZ. In both cases, we were starting from scratch and I was the technical leader of both, I guess, in a group of maybe 30 in this area, led by VP Dr. Yoondong Park (who had maybe 60 people under him incl. other subjects). A short time after starting, the financial crisis hit and the QIS-related work was terminated, and we just did TOF work, culminating in an RGBZ sensor presented at ISSCC in 2011. So the effort on QIS was tiny and short. There was one patent family issued eventually on the SEFET (single-electron, or affectionately, Samsung Electronics, FET). See US 8,546,901 and US 8,803,273.
      I probably cannot say much more about this interesting start just out of respect for Samsung. (Probably my NDA has expired.) But it was a brief and fun exploration.

      At Dartmouth we started fresh and took a different approach because I did not want to start with Samsung IP (and besides, those patents weren't public domain until 2013/2014.)

      Interest in QIS at Samsung has peaked and ebbed a couple of times. Samsung supported some QIS-related work at Dartmouth under their GRO program which will be published in JSSC soon, I expect, but its focus was on low bit depth low power ADCs not pixels and devices were made at TPSco using their global shutter technology. (TPSco's GS pixel worked well, btw, but the CG is way too low for photon-counting applications, as we knew when we started.) That GRO work was performed by my graduate student Zhaoyang "Yang" Yin.

    3. Thank you Eric. I am also quite pleased with this paper. The reconstruction algorithm is actually analogous to the standard ones used in CIS -- just replace the old combination weights by the new ones. Everything else remains linear.


All comments are moderated to avoid spam and personal attacks.