DETECTION OF CELL-MEMBRANES IN HUMAN CORNEAL ENDOTHELIAL MICROGRAMS USING MATHEMATICAL MORPHOLOGY

We propose a method for computer-assisted detection of cell membranes in human corneal endothelial micrograms. Detection of the cell boundary is generally achieved by manually tracing the cell membranes in the micrograms. However, manual tracing is subjective and time consuming. Our method uses modified morphological filters to smooth out noise while keeping linear structures in the image. The processed image is then binarized by template matching. Since the resultant membranes in the image normally consist of broken lines, they must be connected. The identical morphological operation is used again to connect the broken lines. The shapes of cells derived by this method overlap well with those defined by manual tracing.