Illumination Invariant Automated Drop Size Estimation

Estimation of drop size distribution is important to understanding chemical processes and mass transfer operations involving liquid–liquid two-phase flows. This necessitates acquisition of images of fluids under reaction, processing them for identification of drops in images, and estimating their distribution. This process is often automated with advanced image processing algorithms. A novel image processing algorithm for the detection and size estimation of drops from images acquired under varying illumination is reported in this paper. The algorithm is demonstrated to perform under different illumination conditions and image qualities, including varying the image contrast. It uses techniques from 2D stationary wavelet transform, normalized cross-correlation, k-means clustering and transforms each image to a “uniform illumination” domain before identifying and sizing individual drops. The algorithm is unique in its ability to adjust images to a consistent lighting level. It has been tested on 125 different images, analyzing over 9,700 drops, and has achieved a Sauter mean diameter estimation accuracy of 98.85%.