Working with visual data.
WHAT AN IMAGE IS, COMPUTATIONALLY
A two-dimensional array of values, with channels for colour.
WHAT COLOUR SPACES ARE
Different representations: additive primaries, perceptual models, separated brightness and colour.
WHY THAT MATTERS
Some operations are far simpler in one space than another.
WHAT POINT OPERATIONS DO
Transform each pixel independently: brightness, contrast, thresholding.
WHAT SPATIAL FILTERING DOES
Computes each output pixel from a neighbourhood.
WHAT THAT ENABLES
Blurring Sharpening Edge detection Noise reduction
WHAT EDGE DETECTION FINDS
Locations of rapid intensity change, which frequently correspond to object boundaries.
WHAT MORPHOLOGICAL OPERATIONS DO
Modify shapes: removing small features, closing gaps, extracting structure.
WHAT SEGMENTATION DOES
Divides an image into meaningful regions.
WHAT FEATURE DETECTION FINDS
Distinctive points that can be matched between images.
WHAT THAT ENABLES
Stitching, tracking, and reconstructing three-dimensional structure.
WHAT LEARNED APPROACHES CHANGED
Classification and detection, where they substantially exceed hand-designed features.
WHAT CLASSICAL METHODS REMAIN BEST FOR
Well-defined geometric and measurement tasks, with no training data required.
