Multiscale transform has become a key ingredient in many data processing tasks. With technological development, we observe a growing demand for methods to cope with nonlinear data structures such as manifold values. In this paper, we propose a multiscale approach for analyzing manifold-valued data using pyramid transform. The transform uses a unique class of downsampling operators that enables non-interpolatory subdivision schemes as upsampling operators. We describe this construction in detail and present its analytical properties, including stability and coefficient decay. Next, we numerically demonstrate the results and show the application of our method for denoising and abnormalities detection. IntroductionMany modern applications use manifold values as a primary tool to model data, e.g., [13,31,35]. The manifold expresses both a global nonlinear structure with constrained, high-dimensional elements. The employment of manifold values as data models raises the demand for computational methods to address fundamental tasks like integration, interpolation, and regression, which become challenging under the manifold setting, see, e.g., [1,2,24,41]. We focus on constructing a multiscale representation for manifold values using a fast pyramid transform.Multiscale transforms are standard tools in signal and image processing that enables a hierarchical analysis of an object mathematically. Customarily, the first scale in the transform corresponds to a coarse representation, and as scales increase, so are the levels of approximation [33]. The pyramid transform uses a refinement or upsampling operator together with a corresponding subsampling operator for the construction of a fast multiscale representation of signals [6,39]. The simplicity of this powerful method opened the door for many applications. Naturally, recent years found generalizations of multiscale representations for manifold values as well as manifold-valued pyramid transforms [37]. Contrary to classical, linear settings, where upsampling operators are often linear and global, e.g., polynomial interpolation, refinement operators to manifolds values are mostly nonlinear and local operators. One such class of operators arises in subdivision schemes.Subdivision schemes are powerful yet computationally efficient tools for producing smooth objects from discrete sets of points. These schemes are defined by repeatedly applying a subdivision operator that refines discrete sets. The subdivision refinements, which serve as upsampling operators, give rise to a natural connection between multiscale representations and subdivision schemes [4, Chapter 6]. In recent years, subdivision operators were adapted to manifold data and nonlinear geometries by various methods, and so are their induced multiscale transforms, see [39] for an overview.
Multiscale transforms have become a key ingredient in many data processing tasks. With technological development we observe a growing demand for methods to cope with nonlinear data structures such as manifold values. In this paper we propose a multiscale approach for analyzing manifold-valued data using a pyramid transform. The transform uses a unique class of downsampling operators that enable a noninterpolating subdivision schemes as upsampling operators. We describe this construction in detail and present its analytical properties, including stability and coefficient decay. Next, we numerically demonstrate the results and show the application of our method to denoising and anomaly detection.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.