This paper presents a robust image processing technique for bubbly flow measurement over a wide range of void fractions. The proposed algorithm combines geometrical, optical and topological information recorded with high speed cameras to separate and reconstruct the overlapping bubbles. The common difficulties such as overlapping, irregular bubble shape, surface deformation and large clustering in digital image processing are solved by combining different information based on a preset decision table and flow chart. Test with synthetic bubble images is performed to evaluate the reliability of the algorithm and quantify the uncertainty of the data. The result shows that the proposed algorithm can accurately measure bubbly flows with void fraction up to 18% for large bubbles. Four runs of bubbly flow images in a 30 mm x 10 mm rectangular channel are then recorded by three high speed cameras. The area averaged void fraction of these test runs range from 2.4% to 9.1%. The axial and lateral distributions of bubble number density are obtained by the present algorithm for studying the characteristics of these flows. (C) 2016 Elsevier Ltd. All rights reserved.