Differentiation of digital tb images using texture analysis and rbf classifier

In this work, differentiation of positive and negative images of Tuberculosis (TB) sputum smear has been attempted using statistical method based on Gray Level Co-occurrence Matrix (GLCM). The sputum smear images (N=100) recorded under standard image acquisition protocol are considered for this work. Second order statistical texture analysis is performed on the acquired images using GLCM method and a set of nineteen features are derived. Principal Component Analysis (PCA) is then employed to reduce feature sets, to enhance the efficiency of differentiation and to reduce the redundancy. These feature sets are further classified using Radial Basis Function (RBF) classifier. Results show that GLCM is able to differentiate positive and negative TB images. Correlation is found to be high for many of the parameters. Application of PCA reduced the number of features to four which had maximum magnitude in the first principal component. Higher classification accuracy is achieved using RBF classifier. It appears that this method of texture analysis could be useful to develop automated system for characterization and classification of digital TB sputum smear images.