Wall-adherent cells segmentation based on cross-entropy and watershed transform

The microscopic image processing technology is a new solving approach to segment and count wall-adherent cells in anti-virus experiment in vitro. But the segmentation is very stubborn because of the cellspsila multiformity.This paper presents a segmentation strategy based on cross-entropy and watershed transform to segment and count the wall-adherent cells. Firstly, top-hat transform is used to enhance the original cells microscopic image. Suppose the conditional distributions of object and background are modeled with normal distributions, maximum between-class cross-entropy threshold segments the image into binary one. Then morphological filters reduce the burrs and holes in binary image and watershed transform further segments the cells by single-pixel wide edges. Finally, the cells are counted by labeling them.The experiments show that this strategy is effective, simply and strongly adaptive to lighting. The segmentation boundaries are continuious and the cellspsila shapes are well kept.