Evaluation of autofocus algorithms for tuberculosis microscopy

Direct sputum smear microscopy is the most cost-effective method of screening suspected cases of tuberculosis (TB) in high-prevalence countries. Automated microscopy of sputum smears to detect TB would enable rapid screening of large numbers of samples. We evaluate autofocus algorithms for TB microscopy as a step in the development of an automated microscope. Three different focus measures (namely the energy of the image Laplacian, the variance of the log-histogram and the first order Gaussian derivative) are tested on sequences of images of sputum smear slides with varying degrees of content density. A combination of search methods is applied in conjunction with the focus measure to find the position of optimal focus.