Endothelial Cell Image Enhancement using Non-subsampled Image Pyramid

A corneal endothelial cell image provides vast amount of information about a human eye. The cell density and cell shape parameters of a given endothelial cell image help opthamologists in making many vital clinical decisions. The acquired endothelial image is poor in contrast where cell boundaries are masked in the background. Previously, most of the work was based on morphological operations in spatial domain. However, if we think of cell structure as texture hidden in noisy background, we can get help from texture segmentation, a well studied area. In this work, we propose a non-subsampled Gaussian pyramid decomposition of lowpass region. At certain level of the pyramid, we start observing cleaner cell boundary structure which greatly facilitates its segmentation. Once segmented automatic cell counting can be used and simulation results have shown improvement in cell density count