Estimation of camera's noise by uniform target segmentation

This paper proposes a method for evaluating digital camera noise. The method allows to estimate all four noise components: light and dark fixed-pattern (spatial) and temporal noise. We need to register only two scenes: dark and uniform light. Dark spatial and dark temporal noise is measured from two dark frames. Light temporal and light spatial noise is estimated from two light frames using automatic segmentation of a uniform target (ASUT). Noise of visible range (silicon) cameras of four types was measured: consumer Canon EOS M100, machine vision PixeLink PL-B781F, microscope Retiga R6 and scientific MegaPlus II ES11000. Obtained noise values are equal to that ones for the EMVA Standard 1288 within the value spread. The difference is 0.1% (dark temporal), 1.0% (dark spatial), 1.3% (light spatial), and 2.5% (light temporal) only. The proposed method provides fast and accurate camera noise estimation. This can be used to improve information quality and speed of light registration systems.