Feature Extraction and Selection for Cervical Cancer Diagnosis

Abstract

Feature extraction and selection is an important procedure in cell image quantitative analysis and automatic recognition. In this paper, four kinds of features, morphological features, chromatic features, optical density features and texture features are extracted over cell body area, or nucleus area or cytoplasm area and 87 features in all are extracted. Considering the correlation and redundancy of selected features, we propose genetic algorithm using expression of larger between-class scatter and smaller within-class scatter as fitness function to evolve the optimal individual, and 35 features are selected as optimal features to do further cell classification.