Enhanced Skin Lesion Dataset for Classification Tasks
Published: 14 January 2026| Version 3 | DOI: 10.17632/c8wkdtmpjj.3
Contributor:
Aminur SarkerDescription
The Skin2025 dataset was created using skin lesion images from the ISIC Archive. It includes two main classes: benign and malignant, each containing 10,000 images. Several preprocessing steps were applied to improve image quality without removing any samples. Hair artifacts were removed using a black-hat morphological operation and inpainting. CLAHE was used to enhance local contrast, and a bilateral filter reduced noise while keeping edges clear. Finally, automatic white balance correction based on the Gray World method made the colors look natural. The dataset was divided into training, validation, and testing sets. Skin2025 offers a clean and balanced collection of dermoscopic images to support skin lesion classification research.
Files
Institutions
- Daffodil International University
Categories
Dermatology, Image Processing, Medical Image Processing, Deep Learning