Recognition of highly overlapping ellipse-like bubble images

Abstract

This study describes a robust bubble image recognition algorithm that detects the in-focus, ellipse-like bubble images from experimental images with heavily overlapping bubbles. The principle of the overlapping object recognition (OOR) algorithm is that it calculates the overall perimeter of a segment, finds the points at the perimeter that represent the connecting points of overlapping objects, clusters the perimeter arcs that belong to the same object and fits ellipses on the clustered arcs of the perimeter. The accuracy of the algorithm is studied with simulated images of overlapping ellipses, providing an RMS error of 0.9 pixels in size measurement. The algorithm is utilized in measurements of bubble size distributions with a direct imaging (DI) technique in which a digital camera and a pulsed back light are used to detect bubble outlines. The measurement system is calibrated with stagnant bubbles in a gel in order to define the bubble size dependent effective thickness of the measurement volume and the grey scale gradient threshold as a focus criterion. The described concept with a novel bubble recognition algorithm enables DI measurements in denser bubbly flows with increased reliability and accuracy of the measurement results. The measurement technique is applied to the study of the turbulent bubbly flow in a papermaking machine, in the outlet pipe of a centrifugal pump.

The dual effect of viscosity on bubble column hydrodynamics

“…The number of bubbles to be sampled to achieve a reliable BSD is a matter of discussion (Honkanen et al, 2005). Various studies have sampled different numbers of bubbles-between 50 and 100 (Lage and Espósito, 1999), 200 (Wongsuchoto et al, 2003), 250 (Rakoczy and Masiuk, 2009), 300 (Hanselmann and Windhab, 1998), between 250 and 300 (Aloufi, 2011) and 100 (Passos et al, 2015).…”

Comprehensive experimental investigation of counter-current bubble column hydrodynamics: Holdup, flow regime transition, bubble size distributions and local flow properties

“…The number of bubbles to be sampled to achieve a reliable BSD is a matter of discussion in the literature (Honkanen et al, 2005). Various studies have sampled different numbers of bubblesbetween 50 and 100 (Lage and Espósito, 1999), 200 (Wongsuchoto et al, 2003), 250 (Rakoczy and Masiuk, 2009), 300 (Hanselmann and Windhab, 1998), between 250 and 300 (Aloufi, 2011) and 100 (Passos et al, 2015).…”

Laser marked shadowgraphy: a novel optical planar technique for the study of microbubbles and droplets

“…setting a threshold on the minimum acceptable magnitude of the shadow edge gradient. Nevertheless, these criteria are not rigorous and should be carefully calibrated for every set-up and bubble size (Honkanen et al (2005)) to investigate the corresponding measurement volume (and defocus bias error). In addition, due to limitations imposed by the depthof-field of each camera, these criteria are not able to select an arbitrarily small measurement volume and hence, one cannot limit the defocus bias error to a given target uncertainty.…”

“…the Circle Hough transform (e.g. Minor and Sklansky (1981)), the Gradient Pair Method that is used in the present work (Rad et al (2003)) or the curvature-profile method proposed in Pla (1996) and Honkanen et al (2005). Once the bubble is detected, a check for the presence of the two (correct) glare points is performed.…”