A Deep Learning Approach to Ringworm and Vitiligo Identification in Skin Imagery

Diagnosing skin illnesses such as vitiligo and ringworm can be difficult because of their similarity to other dermatological conditions and their different looks. The suggested approach combines Local Binary Patterns and Grey Level Co-occurrence Matrix, two potent feature extraction techniques. While Local Binary Patterns depicts the local texture patterns inside an image, Grey Level Co-occurrence Matrix analyses the spatial relationships between pixels to collect textural information. These traits are successfully used to depict the feature extraction with the unique qualities of vitiligo and ringworm lesions with accuracy $98 \%$, having been retrieved from dermatological photographs. However, previous machine learning research produced inaccurate results. Then, for classification we tried different deep learning models like Google Net, Mobile Net, ResNet50, ResNet101, Xception, CNN, and VGG19. Among these, we found that VGG19 provided good results for us with an accuracy of about $99.19\%$. The pre-trained VGG19 model is improved by utilising deep learning and the extracted characteristics to reliably identify and categorise vitiligo and ringworm lesions. The model’s discriminative power is increased by integrating Grey Level Co-occurrence Matrix and Local Binary Patterns features, and it maintains strong classification performance with less training data by utilising VGG19. All things considered, the suggested method presents a User Interface for the automated diagnosis of vitiligo and ringworm, giving medical practitioners a useful instrument for early identification and treatment, eventually enhancing patient outcomes and care quality.