Hair Removal Methods in Skin Cancer Images Using Black Top Hat Transform and Wavelet Transform

Skincancer is one of the most common forms of cancer, primarily caused by excessive exposure to ultraviolet radiation. (UV). One of the early detection efforts for skin cancer uses a classification model built with a machine-learning approach based on skin cancer images. One of the characteristics of skin cancer images is the presence of hair in the lesion area, which can affect the accuracy of the classification model. In this study, a hair removal method for skin cancer images is proposed using Black Top Hat Transform and Wavelet Transform. The proposed method aims to improve the accuracy of skin cancer image classification models in the Benign, Carcinoma, and Melanoma classes using Convolutional Neural Network models. (CNN). The architecture of the CNN models used are MobileNetV2, MobileNetV3, and VGG19. The research results show that the proposed method can remove hair from skin cancer images while preserving the lesion area. There was an increase in the accuracy of the CNN model on image data resulting from the proposed method by $\mathbf{6 - 2 0 \%}$ for all three architectures. The highest accuracy was obtained from the CNN MobileNetV3 model architecture, with a training accuracy of $\mathbf{9 2 . 4 6 \%}$ and a validation accuracy of $\mathbf{9 0 . 3 3 \%}$.