Local and Non-Local Means Based Mixed Filtering for Video Sequences Denoising

The classical non-local means (NLM) filter can remove Gaussian noise and protect the texture information effectively when the image corruption degree is relatively low. But when the noise in the image is strong, NLM will produce fictitious texture information, which has the disadvantageous influence on its denoising performance. To address this problem, we introduce the local and non-local means based mixed filtering algorithm. This algorithm introduces a threshold into the NL means for pixel pre-selection and the local weighted mean filter is adopted to suppress the fictitious textures. Experiments are carried out on standard video sequences corrupted by Gaussian noise with different standard deviations. Results show that our mixed filter outperforms the classical NLM filter in terms of image smoothing and image denoising when the image is heavily corrupted.