Gaussian noise level estimation in SVD domain for images
Date of Issue2012
IEEE International Conference on Multimedia and Expo (2012 : Melbourne, Australia)
School of Computer Engineering
Accurate estimation of noise level is of fundamental interest in a wide variety of vision and image processing applications as it is critical to the processing techniques that follow. In this paper, a new, effective noise level estimation method is proposed based on the study of singular values of noise-corrupted images. There are two major novel aspects of this work to address the major challenges in noise estimation: 1) the use of the tail of singular values for noise estimation to alleviate the influence of the signal on the data basis for the noise estimation process, 2) the addition of known noise to estimate the content-dependent parameter, so that the proposed scheme is adaptive to visual signal and therefore it enables wider application scope of the proposed scheme. The analysis and experiments results demonstrate that the proposed algorithm can reliably infer noise levels and show robust behavior over a wide range of visual content and noise conditions, in comparison with the relevant existing methods.
DRNTU::Engineering::Computer science and engineering