Adaptive Direction-Guided Structure Tensor Total Variation

TL;DR

This study designs an alternative to STV, which encodes the first-order information within a local neighborhood under the guidance of spatially varying directional descriptors, and proposes an efficient preprocessor that captures the local geometry based on the structure tensor.

Abstract

Direction-guided structure tensor total variation (DSTV) is a recently proposed regularization term that aims at increasing the sensitivity of the structure tensor total variation (STV) to the changes towards a predetermined direction. Despite of the plausible results obtained on the uni-directional images, the DSTV model is not applicable to the multi-directional images of real-world. In this study, we build a two-stage framework that brings adaptivity to DSTV. We design an alternative to STV, which encodes the first-order information within a local neighborhood under the guidance of spatially varying directional descriptors (i.e., orientation and the dose of anisotropy). In order to estimate those descriptors, we propose an efficient preprocessor that captures the local geometry based on the structure tensor. Through the extensive experiments, we demonstrate how beneficial the involvement of the directional information in STV is, by comparing the proposed method with the state-of-the-art analysis-based denoising models, both in terms of restoration quality and computational efficiency.

Figures (6)

• Figure 1: Exemplary outputs of the DPE procedure. The colormaps respectively show (a) the coherence field of unsmoothed $\textbf{g}_L$, (b) its TV regularized version, (c) the final weights, and (d) the final directions for the noise level $\sigma_\eta = 0.15$.
• Figure 2: A rough illustration of the DPE procedure vs. ADSTV denoiser.
• Figure 3: Thumbnails of the grayscale and color images used in the experiments. From left to right and top to bottom: Monarch, Lena, Parrot, Barbara, Fingerprint, Shells, Feather, Plant, Spiral, Chateau, Indigenous, Dog, Zebra, Workers, Swimmer, Rope, and Corns.
• Figure 4: The detail patches showing grayscale image denoising results. The noisy Barbara ($\sigma_\eta = 0.1$), Feather, Lena ($\sigma_\eta = 0.15$), and Fingerprint ($\sigma_\eta = 0.2$) images are restored by using TV (col-1), EADTV (col-2), STV (col-3), NCDR (col-4), and ADSTV (col-5) regularizers. The quantity pairs shown at the bottom of each image are corresponding to the (PSNR, SSIM) values.
• Figure 5: The detail patches showing color image denoising results. The noisy Indigenous, Dog ($\sigma_\eta = 0.15$), and Rope, Spiral, Corns, Chateau, Indigenous ($\sigma_\eta = 0.2$) images (col-2) are restored by using STV (col-3) and ADSTV (col-4) regularizers. The quantity pairs shown at the bottom of each image are corresponding to the (PSNR, SSIM) values.