MODELING OF SURVEILLANCE VIDEO NOISE

摘要

This paper aims at breaking new ground in modeling and estimation of recording noise of surveillance video for further development of new techniques to restore video images. In order to tackle the video denoising problem with non-stationary image contents and various noise sources, a critical task is estimation of varieties of noise in video signals. The estimation is based on a new general integrated surveillance video noise model (GISVNM), which integrates all typical realistic noise models, including signal independent noise model and signal dependent model to model behaviors of Poisson, additive and impulse noises. In particular, the parameters of the Poison and Gaussian based noise models are estimated by using spatial-temporal noise characteristics of the static background of surveillance video, and the parameters of the impulse model are estimated by geometric properties based on spatial characteristics of the video. The experiments showed promising results obtained using the proposed techniques.