Classification of Skin Disease Image using Texture and Color Features with Machine Learning Techniques

Abstract

Skin disease is a serious condition caused by DNA that can lead to death. This damaged DNA begins to develop uncontrollably in cells, and it now multiplies rapidly. Digitalized analysis of malignancy in skin ulcer images is being investigated. However, analysing these images is difficult due to various distracting variables such as reflections on the skin surface.  Lesions of various shapes and sizes, as well as variations in color light.  As a result, verification is done automatically. Recognizing skin disorder is important for pathologists to develop accuracy and skill early on.  In this paper present  a convolutional neural network model built from Deep Learning and compare it to certain machine learning tools for accurate classification of the following seven categories of skin diseases (Akiec,Bcc,Bkl, Mel, Nv, Df,Vasc). To get the graylevel features, we first used image resizing and then transformed RGB to Grayscale image. Second, apply a filter to remove undesired artifacts and noise.  Finally, averaging the input images and extracting texture feature(GLCM,LBP) and color feature (variance, Entropy) characteristics aids in classification accuracy.  The CNN model  gives  84% outperformed other Machine Learning models such as DT, SVM, KNN, and LGBM.