Automating microscope colour image analysis using the Expectation Maximisation algorithm

Dyed barley cells in microscope colour images of biological experiments are analysed for the occurrence of haustoria of the powdery mildew fungus by a fully automated screening system. The region of interest in the images is found by applying Canny’s edge detector to the hue channel of the HSV colour space. Potential haustoria regions are extracted in RGB colour space by an adaptive Gaussian mixture classifier based on the Expectation Maximisation (EM) algorithm. Since the classes cell and haustorium are at very close quarters, their correct separation is a crucial part and needs a constraining mechanism which ties the EM algorithm to its initialisation data to prevent a too large deviation from it.

Cite

Citation style:
Could not load citation form.