The project aims to demonstrate the use of Computer Vision to assist diagnosis by classifying whether a skin lesion is malignant vs benign.

  • Develop image processing pipeline to extract global features (shape, texture, color descriptors) and local features (visual bag of words with keypoint descriptors) of skin lesion images
  • Train machine learning models and implemented model stacking for skin lesion image classification
  • Implement convolutional neural networks and transfer learning to improve the classification accuracy