On the Direction Guidance in Structure Tensor Total Variation Based Denoising

Demircan-Tureyen E., Kamaşak M. E.

9th Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA), Madrid, Spain, 1 - 04 July 2019, pp.89-100 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1007/978-3-030-31332-6_8
  • City: Madrid
  • Country: Spain
  • Page Numbers: pp.89-100
  • Istanbul Kültür University Affiliated: Yes

Abstract

This paper introduces a new analysis-based regularizer, which incorporates the neighborhood-awareness of the structure tensor total variation (STV) and the tunability of the directional total variation (DTV), in favor of a pre-selected direction with a pre-selected dose of penalization. In order to show the utility of the proposed regularizer, we consider the problem of denoising uni-directional images. Since the regularizer is convex, we develop a simple optimization algorithm by realizing its proximal map. The quantitative and the visual experiments demonstrate the superiority of our regularizer over DTV (only for scalar-valued images) and STV.