In order to make the complementary metal-oxide-semiconductor (CMOS) sensor with many advantages more suitable for spatial high resolution imaging, and look for new imaging techniques, a new algorithm called rolling time delay and integration (TDI) in digital domain is proposed, which is better for low light level and push broom imaging. Then the noise sources and their characteristics are analyzed in detail to study the CMOS imaging system′s quality, and a mathematical model of the relation between the digital integrated images′ signal-to-noise ratio (SNR) and integration stages is established. The influence of integration time and illuminance on SNR is dissussed. Finally, validation experiments are carried out using the designed IBIS5-B-1300 TDI CMOS principle prototype. Experimental results show that the proposed algorithm could improve image quality significantly, where the image′s SNR after integrating 10 stages by the proposed algorithm increased to 29.21 dB from the non-integration′s 19.07 dB, and the SNR increased with the stage′s increasing. Theoretical analysis and experimental validation both show that the rolling TDI in digital domain can improve the image SNR by M(σAD+σCMOS)/(MσAD+MσCMOS) times, and the parameters of σAD and σCMOS are related to the selected CMOS sensor, what′s more, σCMOS is effected by the integration time and illuminance.