Skin cancer is one of the dangerous diseases. This study proposes an automated skin lesion detection and classification system leveraging image processing and deep learning techniques. The methodology includes image acquisition, preprocessing (noise reduction, contrast enhancement, and segmentation), and feature extraction using both handcrafted methods (color, texture, and shape analysis) and deep learning-based approaches (CNNs, Vision Transformers), which is challenging in extracting only particular features. This challenge places an important observation in classifying and extracting features. For classification, deep learning architecture ResNet-50 is employed to differentiate between benign and malignant lesions to classify as skin cancer. Results show that deep learning models surpass traditional approaches in achieving superior accuracy with 89.63 for precision and 91.5 score with F1 and 93.6% with accuracy, sensitivity, and robustness in skin cancer detection. However, challenges such as dataset bias, model interpretability, and real-time deployment remain.