Smoker's Tongue Recognition System based on Spectral and Texture Features using Visible Near-Infrared Imaging

Generally, a smoker and nonsmoker tongue could be differentiated through a visual observation performed by medical practitioners. In this research, the smoker's tongue recognition system was developed using a hyperspectral imaging technique with a spectral range 400 to 1000 nm. Smoker tongue profiles were extracted based on a combination of spectral and texture features. The spectral features were computed based on the averaging reflectance area of the posterior part (middle of the tongue) with the size of 64 × 64 pixels. The texture characteristics were extracted as energy, homogeneity, correlation, and contrast from the reflectance area. Both profiles are combined as an input of the PLS (Partial Least Square) method to select the proper feature. An SVM (Support Vector Machine) method was used to classify the data into two classes, i.e., smoker and nonsmoker. Fourty-five respondents were used as a sample and evaluated by the confusion matrix to measure the performance of the proposed system. Based on the experiment, the best combination features are averaging reflectance and correlation, with accuracy obtained from classification and evaluation result of the SVM method is 98.9%. The result shows that the tongue's recognition system that has been built provide the proper classification performance.