Feature Extraction vs Fine-Tuning

Feature extraction freezes layers; fine-tuning updates pretrained weights.

How does feature extraction differ from fine-tuning a pretrained model?