EVALUATION OF BACKGROUND SUBTRACTION EFFECT ON CLASSIFICATION AND SEGMENTATION OF KNEE THERMOGRAM

Presence of inflamed region in knee thermogram indicates the existence of abnormality related to inflammatory diseases. Classification based accurate decision making for knee abnormality and spread of disease analysis is negatively influenced by the unwanted background region of thermogram. In this paper, the effectiveness of background subtraction on knee thermogram classification and segmentation is highlighted depending on accuracy and quantification of segmentation output. The method mainly contains three steps: background subtraction, classification, and segmentation of inflamed region. Automated subtraction of background region is performed using Expectation Maximization method. Extracted features from 50 healthy and 50 arthritis affected knee thermograms are fed to classifiers for accuracy analysis before and after background subtraction. Effect of background subtraction increases the accuracy rate of classification at least by 10%. Popular state of the art segmentation techniques also shows that subtraction of background region enhances the accuracy of knee thermogram segmentation at least by 0.1%.