Present work describes the development of a non-intrusive image analysis technique to measure the bubble size distribution in bubbly flows. Generally image analysis methods for bubble characterization in bubbly flows are suitable at low void fractions. These methods depend mainly on the appropriate thresholding/conversion of the image into a binary image, dividing liquid and bubble objects. Clusters of overlapping bubbles are discarded using the criteria of a shape factor, thus leaving only solitary bubbles within an image. When the void fraction is low, the amount of discarded bubble data is very small compared with the obtained solitary bubble data. For bubbly flows with larger void fractions, the discarded data of overlapping/clustering bubbles is considerable. In this work the overlapping clusters are passed through a watershedding algorithm in order to segment the groups of bubble object areas into individual bubbles. The developed image algorithm is verified with synthetic bubble images and applied for an actual system involving bubbly flow in a pseudo-2D bubble column.