In order to tackle the recording video denoising problem with non-stationary image contents, a critical task is to estimate statistical properties of the composite noise in video signals. After investigation of a variety of test video sequences, based on the wavelet decomposition of the noise, the estimation can be conducted by empirical modeling of the marginal distributions of wavelet coefficients as a subband-dependent parameterized generalized Gaussian distribution. Further development of a new spatio-temporal based composite noise suppression technique is also provided to restore surveillance video, based on the estimation of the parameters of the model. The experiments showed that the proposed techniques obtained promising results.