Skin Texture Analysis Using Machine Learning

According to skin specialist, skin texture has close relation to an individual's health, hormones, hydration, and allergic symptoms. So by procuring one's image texture sample and exposing it to the imaging device we can identify the skin health. Texture analysis is an important tool to analyze the skin texture. The existing means of skin analysis is applicable only for isotropic images. Isotropic images are the ones which is identical in all dimensions. The proposed work includes capturing a skin image and getting its type as output. The steps followed for implementing the proposed work includes acquisition of the skin images then preprocessing them to convert into gray scale and then obtaining various features like mean, standard deviation, skewness and kurtosis using Law's filter and Local Binary Pattern (LBP) feature extraction techniques. After obtaining these features, they are given as input to classifier like Artificial Neutral Network (ANN). The input image set used is of 197 images grouped into training and testing set. The skin images were gathered from various areas of Pune, India and were rated as normal, good or bad skin from a standard dermatologist in the range of 1 to 10. The training phase consists of 70% of the total images and it was tested using remaining 30% images. The accuracy of classifying the images into bad, normal and good skin is observed to be around 99.38%.