Propagation of Noise Statistics in Digital Photofinishing Image Processing

A highly influential factor in the performance of image processing algorithms is the amount of noise present in the digital image. A priori knowledge of the expected levels of noise in the image dramatically improves the performance and efficiency in image processing routines. In a digital photofinishing system, image noise is primarily attributed to film grain and scanner noise. Therefore, if the film and scanner sources are known, it is possible to deduce the expected noise level in a digital image. However, image processing applied to the image after scanning will affect the noise statistics. In order for the image processing algorithms to deliver optimum performance, the estimated noise statistics need to be modified according to each processing step applied. We consider image-processing operations as applying transformations to the image data, and corresponding ones to the image noise statistics. We will discuss analytic equations that approximate the propagation of image noise statistics through several basic image transformations, and their interaction with algorithms in a digital photofinishing image chain.