To be honest, not much covered here that wasn’t covered extensively in Computer Vision.

  • 8.1 explains that images are simply 2d signals
  • 8.2 explains that CNNs generally do a better job than hand-constructed techniques
  • 8.3 gives some examples of basic image processing techniques
  • 8.4 explains stereo image processing
  • 8.5 ML + CV = Best friends

Take home lessons:

  1. Unlike the sensors from Chapter 7, our brains can directly process the 2D information that is captured by a vision sensor. It is difficult to unthink the amount of processing that we perform automatically, augmenting the signal with knowledge and other information that the computer does not necessarily have.
  2. Algorithms described in this chapter aim at reducing information to a lower-dimensional space by removing noise and other spurious information, making the related challenge of understanding the data more tractable.
  3. There is a trade-off between making the data stream more tractable and preserving actual information. As computers and algorithms, in particular machine learning, become more powerful, modern vision systems often blend pre-processing and actual image understanding into a single pipeline.