High-dynamic-range imaging enables recording scenes with both very dark and bright subscenes. When coupled with photon-counting, the highest signal-to-noise ratio is obtained. Solid-state photon-counting can be implemented either with single-photon avalanche diodes (SPADs), or with conventional CMOS image sensors featuring high conversion gain and low readout noise. In SPAD-based imagers, the pixel dead time (Tdead) limits the dynamic range to the maximum counting rate, i.e. 1/Tdead, as well as a nonlinear photon response. Conventional CMOS image sensors with photon-counting capability are limited in dynamic range due to low full-well capacity, while oversampling in time and/or space (like in the quanta image sensor approach) increases the readout noise, thereby deteriorating the photon-counting capability. We present a quantitative analysis on how to use the SPAD photon response nonlinearity and count saturation to actually extend the optical dynamic range far beyond 1/Tdead. Theory and simulations are compared to measurements of the photon response, standard deviation and signal-to-noise ratio for different SPAD recharging (or resetting in CMOS imagers) mechanisms. We also quantify the decrease in signal-to-noise ratio when applying linearization corrections. Results show that by applying active clock recharge, one can extend the optical dynamic range by a factor of 2.8 over 1/Tdead, and by more than 16× over 1/Tdead with active event-driven recharge. It has to be noted that this methodology can be applied to any photon-counting array. Further, we discuss high-resolution image sensor architectures enabling photon counting with single exposure >120dB dynamic range.