Mixed noise face hallucination via adaptive weighted residual and nuclear-norm regularization

Face hallucination has been extensively studied in recent years. The majority of existing approaches operate admirably in noise-free or single camera/atmospheric noise (i.e., Gaussian or impulse) situations. However, when both Gaussian and impulse noises (mixed Gaussian-impulse noise (MIXGIN)) jointly contaminate an image, face hallucination becomes a challenging task. To deal with mixture image corruption, we offer an Adaptive Weighted Residual and Nuclear-Norm Regularization approach in this paper. The method reveals some of the most crucial components of the face hallucination problem. By assigning the weights adaptively to the pixels according to their coding residuals, the proposed method could self-identify the mixed noise and alleviate its effects on the coding process. Moreover, considering the similarity of the training samples and the sparsity of representation coefficients, a nuclear-norm regularization is employed to represent and hallucinate a high resolution face image using correlated and discriminative samples. The method is adaptive to produce suitable coefficients in the presence of mixed noise. To optimize the model, the alternating direction method of multipliers (ADMM) is introduced. The experimental results show that the proposed algorithm gives better performance than the existing face hallucination methods.