Computer Vision

geometric understanding of the world (deep learning for semantic understanding)

2d image → 3d

• Linear Algebra
  • derivatives
• history of cv
  • todo
• pixels and filters
  • linear color spaces
  • image sampling and quantization
  • images as functions
  • filter (kernel)
• systems and convolutions
  • system (cv)***
    • linear shift invariant (LSI)
  • only need to pass impulse to system
• derivatives and edges
  • convolution - filtering
  • cross correlation - compares similarity of two sets of data
  • Edge Detection
  • Filters for Derivatives
• Detecting Lines
  • noise
  • Derivative theorems
  • Sobel Filter
  • Canny edge detector**
  • Hough Transform
  • RANSAC
• Lines and Corners, search
  • Local Invariant Features + Search
  • Harris Corner Detector - REVIEW
• geometry - REVIEW
  • projective space
  • parametrization
  • todo
• detectors and descriptors
  • scale invariant keypoint detection
  • Laplacian of Gaussian LoG
  • SIFT - local detector
  • local descriptor
  • image homographies
• Panorama stitching
  • .
• Global Feature Descriptors
  • Histogram of Oriented Gradients (HoG)
• Current applications - saliency and retargeting
  • image retargeting (ex. responsive UI, adapting films for planes)
  • seam carving
• Motion and Tracking
• Geometric vision, cameras
  • depth estimation, geometric vision
  • pinhole camera
  • calculating vanishing points
  • world and camera coordinates
  • camera intrinsics and extrinsics
• Camera Calibration
  • camera intrinsics and extrinsics
  • estimating camera motion given multiple images
• Structure from Motion
• Segmentation
  • clustering
  • large-scale recognition
• Recognition
  • Dimension-reduction
    • principal component analysis (pca) and singular value decomposition (SVD)
    • Linear Discriminant Analysis (LDA)
• Object Detection
  • gradient descent and backpropagation
  • classifier
  • neural networks

---

basic

• to rotate an image, translate points to desired origin (determine center of image), and apply rotation matrix with theta in radians (to each of the 3 rgb channels)
  • apply inverse rotation matrix if mapping output pixels to input pixels…