Vectorized total variation defined by weighted L infinity norm for utilizing inter channel dependency

Vectorized total variation (VTV) is very successful convex regularizer to solve various color image recovery problems. Despite the fact that color channels of natural color images are closely related, existing variants of VTV can not utilize this prior efficiently. We proposed L∞-VTV as a convex regularizer can penalize the violation of such inter-channel dependency by employing weighted L∞ (L-infty) norm. We also introduce an effective algorithm for an image denoising problem using L∞-VTV. Experimental results shows that our proposed method can outperform the conventional methods.