DISEASE IDENTIFICATION USING FINGER NAIL IMAGE PROCESSING AND ENSEMBLE NEAREST NEIGHBOR CLASSIFIERS OF COLOR FEATURES

Abstract

In the medical domain, image processing plays a vital role in identifying human diseases by inspecting the infected parts of the patient. Usually, human diagnosis is carried out through pathological tests. Those diagnostic methods are invasive. Moreover, the human eye suffers from subjectivity in resolving colors, thus a small variability in nail color may lead to the wrong conclusion. However, computer-assisted diagnosis may detect and recognize such a small change in nail color. Thus this study was to extract color features of fingernail images for identification of normal, anemia, and disease caused by fungus infection using digital image processing techniques and an ensemble of nearest neighbor classifier of color features. The color moments of the diseased and normal nail images were compared and performances of KNN classifiers were evaluated. In this work, the image samples of a total of 150 sample images per hemoglobin blood levels (normal and anemic) and fungus infections were captured by using the smartphone for training, validation, and testing with the proportion of 60%, 20%, and 20%, respectively. The nail portion was segmented and nail color was extracted and combined to form four-color features mean (RGB and HSI), variance (HSI), and range of (HIS), and the color moments of diseased and normal nails were compared to identify diseased and normal nail images. The extracted color features were stored in a vector object and the diseases were identified using built-in ensembles KNN classifiers of color features using MatLab software (2016a). The performance analysis of NIPS-K was done using the statistical measures for binary classification like Sensitivity, Specificity, and Accuracy. The accuracy of classification using color features was 95%, 93%, and 91% for normal, anemic, and fungal infections, respectively