AI Glossary
The complete dictionary of Artificial Intelligence
SIFT (Scale-Invariant Feature Transform)
Scale and rotation invariant interest point detection and description algorithm, using multi-scale spaces and descriptors based on orientation histograms.
SURF (Speeded Up Robust Features)
Accelerated feature detection method using box approximations for Hessian determinant calculation, offering superior performance to SIFT with similar robustness.
ORB (Oriented FAST and Rotated BRIEF)
Binary detector-descriptor combining FAST for detection and BRIEF for description, with added orientation for rotation invariance, offering a free and efficient alternative to SIFT/SURF.
Harris Corner Detector
Classic corner detection algorithm based on analysis of the local gradient autocorrelation matrix, identifying points where intensity variations are significant in multiple directions.
FAST (Features from Accelerated Segment Test)
High-performance interest point detector based on comparing neighboring pixel intensities with a threshold, optimized for real-time computation.
BRIEF (Binary Robust Independent Elementary Features)
Compact binary descriptor generating bit vectors by randomly comparing pixel pairs in a patch, offering extreme speed at the expense of rotation invariance.
HOG (Histogram of Oriented Gradients)
Feature descriptor counting occurrences of gradient orientations in localized portions of an image, particularly effective for object detection.
Binary Descriptor
Compact representation of local features as binary vectors, enabling extremely fast comparisons using XOR operations and Hamming distance.
Feature Matching
Process of identifying corresponding pairs of interest points between two or more images, essential for image stitching, tracking, and 3D reconstruction.
Interest Point
Spatial location in an image presenting a distinctive and repeatable feature, such as a corner, a blob, or a textured region, that can be reliably detected.
Local Descriptor
Numerical vector describing the visual appearance of a region around an interest point, capturing essential information for robust identification despite transformations.
Scale Space
Multi-resolution representation of an image generating versions at different scales to detect features invariant to size changes, fundamental for SIFT and SURF.
RANSAC (Random Sample Consensus)
Robust iterative algorithm estimating the parameters of a model from data containing outliers, widely used to filter incorrect matches in computer vision.
Corner Detector
Class of algorithms identifying points of high curvature in images where gradients show significant variations in at least two orthogonal directions.
Rotation Invariance
Property of a detector or descriptor producing stable results despite image rotations, typically achieved by estimating the local orientation of the patch.
Scale Invariance
Ability of an algorithm to detect and describe the same features regardless of their apparent size in the image, achieved by searching in the scale space.
DoG (Difference of Gaussians) Filter
Operator approximating the Laplacian of Gaussian by subtracting two blurred versions of the image with different standard deviations, used in SIFT for blob detection.
LoG (Laplacian of Gaussian)
Blob detection filter combining Gaussian smoothing with a Laplacian operator, identifying local extremum regions in scale space for scale-invariant detection.
Corner Response Function
Mathematical function quantifying the probability that a pixel is a corner based on the eigenvalues of the local gradient matrix, used in Harris-type detectors.
Non-maximum suppression
Post-processing technique eliminating redundant responses from detectors by keeping only local maxima, ensuring better spatial distribution of interest points.