Object localisation in fetal ultrasound images using invariant features

Abstract

We address the task of object localisation in 2D fetal ultrasound images, where invariance to factors such as image contrast and object orientation is desirable. We build on recent methods for rotation-invariant detection and combine them with oriented measures of image structure derived from the monogenic signal. We test our approach on images containing the fetal heart. Our results suggest that although raw intensity features can achieve robust approximate detection, the structural measures can achieve better localisation.