Jots to Pixels: Image Formation Options for the Quanta Image Sensor

Camera pixels continue to shrink to provide better image resolution, but in doing so they also limit the size of the area designated for light collection. This results in reduced signal-to-noise ratio. A solution to this issue is the concept of the Quanta Image Sensor, a binary, photon-counting image sensor that is oversampled both spatially and temporally. The resulting sensor output is a three dimensional binary cube, which needs to be processed to form a grayscale image. In this thesis, QIS image capture and processing are simulated with MATLAB functions in order to investigate ways the 3D binary data can be manipulated to produce the best quality images.;The simplest way to convert the binary pixel ("jot") data into grayscale pixels is to sum groups of jots together and assign the sum an intensity value. It is shown that this method is highly effective, and the size of the group can be increased for a gain in SNR, though at the cost of resolution. Alternatively, groups of jots can be overlapped and then the result downsampled for further gains in SNR with no loss in resolution. It is also found that the SNR can always be improved by increasing the number of captured frames in the third dimension, although this increases total integration time. The SNR can also be extended over a wide dynamic range by successively decreasing the exposure time per frame before summing jots into pixels.;Finally, results of summing groups of jots are compared to an iterative maximum a posteriori solution, and it is found that the MAP solution could allow for greater output SNR for a small number of exposure frames if iterated for long enough. However, summing jots leads to higher overall SNR in significantly less time for a given number of frames.