Non-convex hybrid total variation for image denoising

Image restoration problems, such as image denoising, are important steps in various image processing method, such as image segmentation and object recognition. Due to the edge preserving property of the convex total variation (TV), variational model with TV is commonly used in image restoration. However, staircase artifacts are frequently observed in restored smoothed region. To remove the staircase artifacts in smoothed region, convex higher-order TV (HOTV) regularization methods are introduced. But the valuable edge information of the image is also attenuated. In this paper, we propose non-convex hybrid TV regularization method to significantly reduce staircase artifacts while well preserving the valuable edge information of the image. To efficiently find a solution of the variation model with the proposed regularizer, we use the iterative reweighted method with the augmented Lagrangian based algorithm. The proposed model shows the best performance in terms of the signal-to-noise ratio (SNR) and the structure similarity index measure (SSIM) with comparable computational complexity.

► We proposed a non-convex hybrid TV regularization method. ► We use the iterative reweighted method with the augmented Lagrangian based algorithm. ► We report the proposed model shows the best performance in term of SNR.