Automatic classification of cervical cells-using the frequency domain

The authors present a report of a preliminary study to investigate the effectiveness of spectral analysis techniques to classify cervical cell images as normal or abnormal. In particular an examination was made of the use of texture in the frequency spectra of single cell images to identify the cell class and thereby to discriminate between normal and abnormal. The frequency spectra representation offers several advantages over the original spatial domain image including data compression and avoiding the need to segment the image. An additional goal was to investigate the contributions made by nuclear shape, nuclear texture and cell texture