Segmentation of textured cell images based on frequency analysis

A novel frequency analysis algorithm for segmentation of textured cells is presented. The algorithm is developed based on an ideal simulation model and is applicable to real cell images. A simulated cell image is assumed to have an ellipse-like region of textured interior embedded in a relatively flat background. The size of the original image is expanded multiple times by extrapolating it to additional regions with estimated background intensities before a larger sized discrete Fourier transform (DFT) is applied. The idealised model for the cell images shows a direct relationship between the boundaries of the cell regions and the inner zero-crossing lines in the large-sized DFT of the expanded images. The shape, size and orientation of the cell region are determined by the parameters derived from the estimated inner zero-crossing line in the DFT whereas the position of the cell region is determined by searching for the location of the minimum in the moving average with the window shaped the same as the previously acquired cell region. Experimental results of both the simulated and the real microscopic cell images are provided to show the performance of the proposed algorithm.