Here we introduce a robust feature extraction process for automatic detection of five classes of lung tissue patterns. The feature vector is constructed considering texture, shape and edge information to face two challenges such as low range inter class divergence and high level intra class divergence. To design feature vector Gabor filter is applied to each image patch, texture features are extracted by Local binary pattern (LBP), gradient features by Histogram of oriented gradients (HOG) and intensity features by Histogram bin. The feature vector thus obtained is labelled with support vector machine (SVM) classifier whose performance is compared with various popular classifies KNN, LDA. The current work is implemented on ILD database which shows biased performance over the state- of -the-arts.
Volume 11 | 03-Special Issue
Pages: 1252-1260