Spatial-frequency domain nonlocal total variation for image denoising (Q2026428)

The authors study image denoising by means of the nonlocal total variation (NLTV) model. For this model, they deduce Stein's Unbiased Risk Estimation (SURE). Moreover, they analyse frequency domain nonlocal total variation (FNLTV) by considering that the Fourier transform structures the image data in a totally different way from the standard basis. Then they obtain a spatial-frequency domain nonlocal total variation model (SFNLTV) by extending the NLTV model in spatial domain and frequency domain. This proves to be better than NLTV for texture and fine details in images.