Automatic Classifying of Cervical Cells Using Fourier Spectral Features

Abstract

A method of automatically classifying cervical cells from Pap smear images using the Fourier Transform-based Features is proposed. To avoid the error occurred from the cell and nucleus segmentation process, we proposed the set of simplified features derived from the two-dimensional Fourier spectrum using the discrete Fourier transform. The features in the proposed method are obtained from the frequency components along the circle of radius centered at the center of the spectrum and the frequency components along the radial line having an angle. Each section of frequency components is divided into subsection. Mean value of each subsection are computed and used as the features for classification. The features are used to discriminate cells as a two-class problem to classify between the normal and abnormal cell. Classification experiments are conducted using 5-fold cross validation. The efficiency of four classifiers including K-nearest neighbor (KNN) support vector machine (SVM), Random Forest, and Adaptive Boosting (AdaBoost) are investigated. The performance of the proposed feature to classify the normal and abnormal cells show promising performance with accuracy of classification from all classifiers is more than 93%. SVM shows the best classification rate at 94.38%.