Also called image recognition. Use Convolutional Neural Networks (CNN).

Pre-processing techniques

• data normalization:
  • convert numerical representation of pixels to a predefined range, e.g. (0, 1) or (-1, 1)
• image augmentation (Regularization#^933b29)
• image standardization
  • resize to make consistent image sizes

Handle imbalance in image data

• merge very similar categories
• resampling (Best Practices & Common Pitfalls in Machine Learning#^5cfbc3)
• adjust the CNN's loss function to wight mistakes on minority categories more than mistakes on majority categories

Avoid overfitting

• image augmentation (Regularization#^933b29)
• dropout (Regularization#^b76d6c)
• early stopping (Regularization#^9ff80e)
• Transfer Learning and Multi-task Learning