A method of eliminating the signal-dependent random noise from the raw CMOS image sensor data based on Kalman filter

Abstract

Many conventional denoising solutions adopt the channel-dependent noise model, which is less suitable than signal-dependent noise model to describe the noise characteristic of CMOS image sensor. In this paper, the cascaded Kalman filter is designed to eliminate the signal-dependent random noise from the raw data of CMOS image sensor. The major contribution of this work is that it establishes the state equation corresponding to the circuit structure of sensor pixel, instead of designing the denoising algorithm only from a mathematical point of view. The experimental results confirm that the performance of the proposed method outperforms the conventional ones.