Image Classification

• One of the most popular tasks in Computer Vision
• Goal: assigning labels to images
• Started before Deep Learning was around
• Traditionally (before DL)
  1. Preprocessing (normalize colors etc)
  2. Extracting featurs (many are gradient-based)
  3. Learning Algorithms (aggregators)
  4. Label Assignment
• Success depends on quality of the Feature descriptor
  • Which had to be hand-engineered, not learned
• With DL: no feature descriptors necessary

Possible problems for Image Classifications: #

• Color Differences
  • Occlusions
  • Background clutter (object color == background color)
• Representation Differences
  • cartoon drawing

See also #

Regression vs Classification