Noise Reduction for Image Signal Processor in Digital Cameras

Most of noise reduction methods assume an additive white Gaussian noise (AWGN) model. However, the real noise is not AWGN. Therefore, such noise reduction methods are not effective in real images captured by digital cameras. In this paper, we propose the noise reduction method for image signal processor (ISP) in consumer digital cameras. To efficiently reduce the noise in ISP, we estimate the Skellam noise model in Bayer domain, and then we adaptively reduce the noise according to intensity values. Our noise model doesn't need to be rebuilt according to input images if it is made once. Thus, the proposed method can be a suitable algorithm for ISP. Moreover, the experimental results show that the proposed method reduces the real noise efficiently while the image detail is preserved.