A double total variation regularized model of Retinex theory based on nonlocal differential operators

Image characteristics, such as texture, edge, smoothness, can be much better preserved by using nonlocal differential operators based on patch-distances in image processing. In this paper, we apply with nonlocal differential operators to some existing variation models of Retinex, such as the nonlocal variation model of Retinex (NL_VR); the nonlocal TV regularized model (NL_TV_R) and the nonlocal total variation regularized model with constraints (NL_TV_C). And then we improve and establish a double total variation regularized model of Retinex theory (DTV) and the nonlocal double total regularized model (NL_DTV), which could handles better edges in the illumination. Experiments show that our proposed method and Split Bregman algorithm presented in this paper have higher computational efficiency and accuracy.