An SVM Based Skin Disease Identification Using Local Binary Patterns

Researches on identification of Skin diseases from the digital images are increasing due to multidimensional challenges of the domain. Most of the researches are based upon the freely available digital images from the internet instead of real ground truth data set. To address these problems, we first created a ground truth dataset consisting of 876 images of human skin affected with three prevalent skin diseases of the Indian subcontinent (viz. leprosy, tineaversicolor and vitiligo collected from the patients) together with normal skin and then developed a mechanism to recognize them automatically. It is worthy to mention here, leprosy, vitiligo (at early stage) and tineaversicolor are hypo pigmenting disorders and very similar in lesion shape and color. All the images are divided randomly into train and test sets, approximately in the ratio 4:1 for each class. For recognition of the diseases from the skin images different popular texture and frequency domain features such as Local Binary Pattern(LBP), Gray Level Co-occurrences Matrix (GLCM), Discrete Cosine Transform (DCT) and Discrete Fourier Transform (DFT) have been used with Support Vector Machines (SVM) based classifiers. Maximum recognition accuracies of 89.65% has been observed on test set using the LBP feature set. To the best of our knowledge this is the first automated noninvasive system to identify the three important skin diseases from digital images of the affected skin regions.