Pengyuan Li, Granite Vision Team
CVPR 2025
Periodogram is an important tool for analyzing time series of mixed spectra that can be decomposed as sinusoids plus noise. While effective in many situations, the ordinary periodogram has two major shortcomings: it cannot resolve sinusoids whose frequencies are separated by less than 1 cycle per unit time; it does not possess sufficient robustness against heavy-tailed noise such as outliers. An alternative periodogram is introduced in this article with the aim of improving the frequency resolution as well as the robustness of the ordinary periodogram. The new periodogram, called bivariate ℓ1 -periodogram, is derived from the maximum likelihood method of multiple frequency estimation under the assumption of Laplace white noise. The desired high-resolution and robustness property of the bivariate ℓ1 -periodogram is confirmed by simulation studies. Superior statistical efficiency over alternative methods is also demonstrated. © 2010 Elsevier B.V. All rights reserved.
Pengyuan Li, Granite Vision Team
CVPR 2025
Massimo Maresca, Hungwen Li, et al.
Machine Vision and Applications
George Saon
SLT 2014
Jinyu Zuo, Nalini K. Ratha, et al.
CVPRW 2008