AI Glossary
The complete dictionary of Artificial Intelligence
Active Contour
Deformable model that evolves under the influence of internal and external forces to adapt to object boundaries in an image.
Snake (Model)
Type of active contour defined by a parametric curve that deforms to minimize a functional energy including regularization terms and data attachment terms.
Level Set
Numerical method for tracking interfaces and shapes, representing the contour as the zero level of a higher-dimensional function.
Gradient Map
Image representing the magnitude of intensity change pixel by pixel, used to locate strong variations characteristic of edges.
Laplacian of Gaussian (LoG)
Filter combining Gaussian smoothing and a Laplacian operator to detect edges by identifying zero-crossings of the second derivative.
Difference of Gaussians (DoG)
Efficient approximation of the Laplacian of Gaussian, obtained by subtracting two smoothed versions of the image to highlight structures at different scales.
Vector Force Field
Two-dimensional map assigning a vector to each pixel, guiding active contours toward object boundaries by combining attraction and repulsion.
Contour Curvature
Measure of the rate of change of direction of a contour, used as a regularization term in active contour models to avoid irregularities.
Stopping Function
Term in active contour evolution equations that slows or stops the movement of the contour when it reaches object boundaries.
Mumford-Shah Model
Variational formulation for image segmentation that simultaneously seeks to segment the image and approximate each segment by a smooth function.
Watershed Segmentation
Morphological method that considers the image as a topographic relief, where contours correspond to ridge lines separating catchment basins.
Geodesic Contour
Variant of active contours where the contour evolves by minimizing a geodesic energy weighted by a stopping map, offering better topological convergence.
Bilateral Filter
Non-linear smoothing filter that preserves contours by weighting pixels both by their spatial distance and intensity similarity.
Deep Learning Edge Detection
Approach using trained convolutional neural networks to directly predict edge maps or boundary probabilities between objects.