Robustness of sift feature descriptors to imaging parameters in laser scanning microscopy

Abstract

Confocal laser scanning microscopy relies on illuminating the specimen with a focused scanning laser beam and constructing sharp optical sections/images of the investigated specimen by allowing signals from the focal plane only via a pinhole. It is a valuable technique to study fluorescent cells and tissues in-vivo. Its power and potential can be augmented by the use of computer vision methods. However such methods are typically transferred to the microscopy field directly without taking microscopy-specific variations into account. Accordingly in this study we evaluate the robustness of SIFT feature descriptors, a popular computer vision method, against variations in microscopy-specific parameters over a benchmark dataset acquired in a principled manner. Results show that SIFT descriptors are highly robust against variations in laser beam power, whereas their robustness diminishes with larger variations in photomultiplier tube gain. © 2018 IEEE.