<p>Melanoma detection at the earliest reduces the mortality rate. There are many clinical and computer-vision techniques available to detect melanoma at its early stages. The increasing population that is affected by melanoma is growing day by day. So it’s difficult for clinical testing by dermatologists as the number increases practically. Image processing techniques play a vital role in melanoma detection. The growing number of healthcare facilities helps humans detect the presence of melanoma with the help of high-resolution medical devices. Therefore, this paper proposes a Federated learning framework using novel image processing techniques at local models. The local model is designed using an image processing technique that contains the CapsNet and ensemble transfer learning for image classification. The CapsNet efficiently detects the presence of melanoma characteristics in the images at various orientations, and ensemble transfer helps to detect future images with the help of knowledge from the previous models and uses the ensemble model to attain higher accuracy. The proposed framework showed 98.8% in terms of accuracy, and local models individually showed 96.4% accuracy.</p>