본문 바로가기

Linear-Map-Based Tensor CUR Decomposition for Robust Video Foreground-Background Separation

문서 유형
연구출판물 · 학술지 논문
등재
SCIEScopus
저널 순위
JCR Q2Scopus Q1
저널
IEEE SIGNAL PROCESSING LETTERS33 · pp.2849 · 2026
피인용
Web of Science 0Scopus 0
FWCI (Scopus)
0
JCR IF
JCR 2024 3.9
ISSN
1070-9908

저자

  1. 로페즈 모레노 수사나제1저자Pusan Natl Univ, Humanoid Olfactory Display Ctr, Dept Math, Busan 46241, South Korea
  2. 이준호Pusan Natl Univ, Humanoid Olfactory Display Ctr, Dept Math, Busan 46241, South Korea
  3. Kim, Taehyeong교신저자Kyungpook Natl Univ, Nonlinear Dynam & Math Applicat Ctr, Daegu 41566, South Korea

초록

Foreground-background separation in surveillance videos can be modeled as the decomposition of a third-order tensor into a low-rank background and a sparse foreground component. This paper proposes a robust tensor column-row (CUR) decomposition under the linear-map-based tensor-tensor multiplication, which supports different linear mapping choices within a unified algebraic formulation. The method selects informative row and column indices by a rank-revealing interpolation rule, performs slice-wise matrix CUR factorizations in the image of a linear transformation, and reconstructs the separated components in the original space through the pseudo-inverse map. Theoretical analysis establishes an exact reconstruction result under a multirank condition and a perturbation bound for additive noise. For video foreground-background separation, the proposed decomposition is embedded in a training-free low-rank plus sparse pipeline using median absolute deviation based thresholding. Experiments on standard background subtraction benchmarks show competitive reconstruction quality for both background estimation and foreground extraction when compared with recent matrix-based and tensor-based decomposition methods.

키워드

Background subtractionrobust decompositiontensor decompositiontensor-tensor multiplicationvideo signal processing

식별자

Scopus EID
2-s2.0-105044315584
WoS UID
001824152700008