Structural Similarity-Optimal Total Variation Algorithm for Image Denoising

摘要

Image denoising is a traditional problem which has been tackled using a variety of conceptual frameworks and computational tools. Total variation-based methods have proven to be efficacious toward solving image noise removal problems. Its purpose is to remove unnecessary detail and achieve optimal performance in terms of mean squared error (MSE), a metric that has been widely criticized in the literature due to its poor performance as an image visual quality assessment. In this work, we use structural similarity (SSIM) index, a more accurate perceptual image measure, by incorporating it into the total variation framework. Specifically, the proposed optimization problem solves the problem of minimizing the total gradient norm of restored image and at the same time maximizing the SSIM index value between input and reconstructed images. Furthermore, a gradient descent algorithm is developed to solve this unconstrained minimization problem and attain SSIM-optimal reconstructed images. The image denoising experiment results clearly demonstrate that the proposed SSIM-optimal total variation algorithm achieves better SSIM performance and better perceptual quality than the corresponding MSE-optimal method.