Weighted averaging-based sensor pattern noise estimation for source camera identification

Sensor pattern noise has been broadly used in the literature for image authentication and source camera identification. The abundant information that a sensor pattern noise carries in terms of the frequency content makes it unique and hence suitable for source camera identification. The traditional approach for estimating the sensor pattern noise uses a set of images to estimate a pattern residual signal from each image. The estimated residual signals are then averaged to obtain the sensor pattern noise. This is based on the assumption that each residual signal is a noisy observation of the sensor pattern noise. Such an assumption is well justified in practice because the images are acquired under different conditions making the corresponding residual signals distinct from each other. For instance bright images provide better sensor pattern noise estimation than dark images. Also, saturated pixels cause undesirable noise in residual signals. Inspired by this observation, a weighted averaging approach is proposed for efficient sensor pattern noise estimation. The proposed approach has been validated with two sensor pattern noise estimation techniques from the literature and significant improvements have been shown through experimental results.