Fast Image Registration Using Cepstral Analysis of Inverted Gradient Distance Maps
Author(s)
Gonzalez, Ruben
Griffith University Author(s)
Year published
2018
Metadata
Show full item recordAbstract
FFT based image registration methods are closed form solutions that are robust to global noise. They are also typically much faster than feature matching approaches. The main appeal of using cepstral analysis for this task is that it requires one third less FFTs compared to phase correlation. However, this advantage is lost in the computation of the logpolar transform needed to recover scale and orientation. This paper proposes a novel, alternate approach using an Inverted Gradient Distance Map that reduces the number of FFTs and computation time required to register images using the Cepstrum to half of that needed in using ...
View more >FFT based image registration methods are closed form solutions that are robust to global noise. They are also typically much faster than feature matching approaches. The main appeal of using cepstral analysis for this task is that it requires one third less FFTs compared to phase correlation. However, this advantage is lost in the computation of the logpolar transform needed to recover scale and orientation. This paper proposes a novel, alternate approach using an Inverted Gradient Distance Map that reduces the number of FFTs and computation time required to register images using the Cepstrum to half of that needed in using Phase Correlation.
View less >
View more >FFT based image registration methods are closed form solutions that are robust to global noise. They are also typically much faster than feature matching approaches. The main appeal of using cepstral analysis for this task is that it requires one third less FFTs compared to phase correlation. However, this advantage is lost in the computation of the logpolar transform needed to recover scale and orientation. This paper proposes a novel, alternate approach using an Inverted Gradient Distance Map that reduces the number of FFTs and computation time required to register images using the Cepstrum to half of that needed in using Phase Correlation.
View less >
Conference Title
2017 INTERNATIONAL CONFERENCE ON IMAGE AND VISION COMPUTING NEW ZEALAND (IVCNZ)
Subject
Information and computing sciences