In this work an image processing technique is proposed to improve the measurement of ellipsoidal ob- jects, such as bubbles in dispersed flows. This novel algorithm devoted to the measurement of bubble size, shape and trajectory is applied to binarised images from a gray-level gradient filter. To improve data statistics, an ellipse fitting method is employed to take into account truncated bubbles at the image edges. Then, an original approach is proposed to enable the segmentation of overlapping bubbles. The complete algorithm is evaluated on synthetic images and on real images for an air-bubble swarm within water. This new and robust methodology enables to increase substantially (more than 40%) the number of bubbles detected and thus to improve data convergence.