CMOS sensors used in high-quality imaging systems of human eye require subtraction of the fixed pattern noise (FPN)
to increase the signal to noise ratio (SNR). In this paper, we evaluated the efficiency of the FPN subtraction as a function
of temperature (in the range from +70°C to -50°C) and integration time (from 12.5 to 287.5 ms) for a low-cost
commercial CMOS sensor. We also compared the quality of the photos taken at different sensor temperatures on a
phantom and on voluntary subjects. The experimental set-up includes a modified fundus camera for retinal examination
(Centervue SpA, Italy) equipped with a 5Mpixel CMOS sensor and a temperature regulation system. The problem of
water condensation on the sensor surface at low temperatures was overcome by using an original sensor sealing method.