Detecting cells in DIC microscope images using a high level Bayesian model and template matching

Identification of objects in differential interference contrast (DIC) microscope images by digital image analysis is a hard task. While DIC microscopy is well suited to visualisation of near-transparent cells, the microscope optics cause a pattern of light and dark cell edges to appear in the image, giving a pseudo 3D effect. A high level Bayesian statistical approach is described as an alternative, for cells of specifiable geometric shape, which has the potential to cope more easily with clustered or overlapping cells. The prior model represents initial knowledge about the objects and/or object configuration by a probability distribution on their parameters, while the image model or likelihood specifies a joint probability function for the grey levels given the objects. Results of the Bayesian method are presented for comparison with those of the full template matching.