Skin cancer is a significant health issue affecting millions annually, often due to lifestyle changes and increased solar exposure. Early detection and staging are crucial due to the high mortality rate associated with skin cancer. This study introduces DVFNet, a deep learning-based method for detecting skin cancer from dermoscopy images. The method pre-processes images using anisotropic diffusion to enhance quality and remove artifacts and noise. It combines the VGG19 architecture and Histogram of Oriented Gradients (HOG) for feature extraction, addressing the problem of imbalanced classes in the ISIC 2019 dataset using SMOTE Tomek. The model segments images to identify areas of damaged skin cells and creates a feature vector by fusing HOG and VGG19 features. The CNN classifies these vectors, achieving an accuracy of 98.32% on the ISIC 2019 dataset. ANOVA statistical tests validate the model's accuracy, making it a valuable tool for healthcare professionals to detect skin cancer early.Skin cancer is a significant health issue affecting millions annually, often due to lifestyle changes and increased solar exposure. Early detection and staging are crucial due to the high mortality rate associated with skin cancer. This study introduces DVFNet, a deep learning-based method for detecting skin cancer from dermoscopy images. The method pre-processes images using anisotropic diffusion to enhance quality and remove artifacts and noise. It combines the VGG19 architecture and Histogram of Oriented Gradients (HOG) for feature extraction, addressing the problem of imbalanced classes in the ISIC 2019 dataset using SMOTE Tomek. The model segments images to identify areas of damaged skin cells and creates a feature vector by fusing HOG and VGG19 features. The CNN classifies these vectors, achieving an accuracy of 98.32% on the ISIC 2019 dataset. ANOVA statistical tests validate the model's accuracy, making it a valuable tool for healthcare professionals to detect skin cancer early.