Robust and discriminative image representation: fractional-order Jacobi-Fourier moments. (July 2021)

• Record Type: Journal Article
• Authors: Yang, Hongying; Qi, Shuren; Tian, Jialin; Niu, Panpan; Wang, Xiangyang
• Is Part Of: Pattern recognition. Volume 115(2021)
• Publication Date: 2021-07
• Subjects: Image representation -- Fractional -- Jacobi-Fourier moments -- Robustness -- Discriminability; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.107898
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We define a new set of generic orthogonal moments, called Fractional-order Jacobi-Fourier Moments (FJFM), which is a generic version of many existing classical and fractional-order moments. We develop a novel framework to improve both the robustness and discrimination power of image global representation, named Mixed Low-order Moments Feature (MLMF), by fully exploiting the time-frequency analysis property of FJFM. Extensive experimental results and a real-world application are given to verify the advantages of the proposed image representation scheme, with respect to robustness and discriminability. Abstract: Robust and discriminative image representation is a long-lasting battle in the computer vision and pattern recognition. Moment-based image representation model is effective in satisfying the core conditions of semantic description, due to its geometric invariance and independence. However, moment-based descriptors suffer from a contradiction between the robustness and discriminability, which limits the further improvement of description quality. In this paper, a set of generic moments along with a novel representation framework are proposed to mitigate this troublesome contradiction. We first define a new set of orthogonal moments, named Fractional-order Jacobi-Fourier Moments (FJFM), which is characterized by the generic nature and time-frequency analysis capability. We then develop a new framework to improve both the robustness and discriminability of image representation, called Mixed Low-order Moment Feature (MLMF), by fully exploiting the time-frequency property of FJFM. Extensive experimental results and a real-world application are provided to demonstrate the superior performance of our FJFM-based MLMF, with respect to robustness and discriminability.