PRNU Estimation of Linear CMOS Image Sensors That Allows Nonuniform Illumination

Abstract

Photo response nonuniformity (PRNU) is a crucial parameter in evaluating and choosing CMOS image sensors. The current methods and standards use the integrating sphere to illuminate the sensor and assume a known uniform illumination. And PRNU is calculated based on the spatial variance of images. In practice, the uniformity of illuminance can be hardly guaranteed. A slight nonuniformity will cause a significant increase in spatial variance and bring unacceptable evaluation errors. In this article, we present a novel method to estimate PRNU of linear CMOS image sensors. The method allows nonuniform illumination and does not require specific settings for the light source and sensors. First, an observation model is established for the data of nonuniformly exposed images. The model is composed of an irradiance model and a noise model. The irradiance model describes the original output, and the noise model depicts the noise variance as a function of local illuminance. PRNU is a parameter in the noise model to be estimated. Then, the model parameters including PRNU are estimated by a two-stage method based on maximum likelihood estimation. We also use simulations to validate the developed methods and demonstrate the robustness of the model under sensor misalignment. Finally, the proposed method is illustrated by a measurement experiment.