image_segmentation.md

Image Segmentation

• Graph theoretic segmentation

  • Normalized Cuts
  • Using textons (texture features)

• Segmentation as Energy Minimization

  • Grab Cut
  • Graph Cuts
  • Markov Random Fields (MRF) & Conditional Random Fields (CRF)

• Semantic Segmentation

  • Why?
    • Satellite imaging
    • Medical imaging
    • Facial detection and recognition
    • Image-based search
    • Retail and fashion
    • Robot vision and understanding
    • Autonomous driving

• Post-processing

  • "They [CRFs] can enforce global and local consistency of the predicted segments." Source: https://www.reddit.com/r/MachineLearning/comments/9e13kg/d_why_fully_connected_crf_help_in_sharp_labelling/
  • "The key idea of CRF inference for semantic labelling is to formulate the label assignment problem as a probabilistic inference problem that incorporates assumptions such as the label agreement between similar pixels." Source: https://www.robots.ox.ac.uk/~szheng/papers/CRFasRNN.pdf
  • "The pairwise energies provide an image data-dependent smoothing term that encourages assigning similar labels to pixels with similar properties." Source: https://www.robots.ox.ac.uk/~szheng/papers/CRFasRNN.pdf

• CRF papers to review

  • https://arxiv.org/pdf/1504.01013.pdf
  • https://arxiv.org/pdf/1503.02351.pdf

• Source: http://vision.stanford.edu/teaching/cs231b_spring1213/slides/segmentation.pdf

• Recurrent Instance Segmentation