Cytoplasm Contour Approximation Based on Color Fuzzy Sets and Color Gradient

Here we propose a method for contour detection of cells on medical images. The problem that arises in such images is that cells' color is very similar to the background, because the cytoplasm is translucent and sometimes overlapped with other cells, making it difficult to properly segment the cells. To cope with these drawbacks, given a cell center, we use hue and saturation histograms for defining the fuzzy sets associated with cells relevant colors, and compute the membership degree of the pixels around the center to these fuzzy sets. Then we approach the color gradient (module and argument) of pixels near the contour points, and use both the membership degrees and the gradient information to drive the deformation of the region borders towards the contour of the cell, so obtaining the cell region segmentation.