Feature detection can use point matching based on feature or boundary matching {feature detection methods}: corner detection, scale-invariant features (SIFT), and speeded-up robust features (SURF). Features are good image-category descriptors.
Algorithms {feature detection algorithms} can detect features.
descriptor
Properties {X-variable, vision} {X descriptor, vision} can describe features and have components.
canonical factor analysis
Factor analysis has a basic method.
centroid method
Factor analysis can use centroids.
Correlation Analysis
Properties and features have relationships.
correspondence factor analysis
Factor-analysis methods can use variable frequencies relative to properties, find chi-square values, and find principal components.
disjoint principal component
Principal components can be independent.
eigenvalue-one criterion
Thresholds can be how many components have eigenvalues greater than one.
eigenvector projection
Unsupervised linear methods can find factors.
Evolutionary Programming
Models can add and subtract randomly selected variables, with crossing-over, and evaluate for "fitness" or best fit. Extinction can happen more or less frequently or in more or fewer species. More-frequent extinctions have fewer species. Values follow power laws, because events can cause few or many extinctions.
evolving factor analysis
Methods can analyze ordered data.
explained variance percentage
Methods can indicate component number required to reach 90% of total variance.
factorial design
Designs can try to ensure design-space sampling, even if one position varies.
Genetic Function Algorithm
Linear property sets can have different values, change values by crossing-over between related genes, and have random changes, to select best fit.
latent variable
Variables can be linear descriptor combinations.
linear discriminant analysis
Supervised methods, in which boundary surface minimizes region variance and maximizes between-region variance, can put compounds into groups by activity level.
linear learning machine
Supervised methods can divide n-dimensional space into regions using discriminant functions.
maximum-likelihood method
Factor-analysis methods can find factors.
multidimensional scaling
Metric or non-metric methods can analyze similarity or dissimilarity matrices to find dimension number and place objects in proper relative positions.
multivariate adaptive regression spline
Non-parametric methods can find factors.
Mutation and Selection Uncover Models
Models can add and subtract randomly selected variables, with no crossing-over, and evaluate for "fitness" or best fit. Low mutation rates allow natural selection to operate on populations to move toward fitter genotypes. Intermediate mutation rates cause population to move toward and away from fitter genotypes. High mutation rates make many genotypes with direction, so high mutation blocks selection processes.
For any mutation rate, if gene number is too great, change rate is too great, and organism becomes extinct {error catastrophe, extinction}. Therefore, gene number has a limit if organisms do not make new species or find new environments.
Perhaps, cells and ecosystems also have upper limits to complexity. Complexity can increase with migration or speciation.
non-linear iterative partial least squares
Unsupervised linear methods can represent data as product of score matrix, for original observations, and loading-matrix transform, for original factors.
non-linear mapping
Topological mapping factor-analysis method uses linear variable combinations to make two or three new variables.
predictive computational model
Property information can predict behavior.
principal-component analysis
Variable principal components can be linear-descriptor combinations. Unsupervised linear methods can represent data as product of score matrix, for original observations, and loading-matrix transform, for original factors. PCA factor-analysis method uses linear variable combinations to make two or three new variables. PCA reduces unimportant variables.
principal-component regression
Singular-value decomposition can find singular-value sets to predict and project regression to latent structures.
principal factor analysis
Modified PCA can find principal factors.
Procrustes analysis
Methods can identify similarity descriptor sets.
QR algorithm
Methods can diagonalize matrices.
rank annihilation
Unsupervised linear methods can find factors.
response-surface method
Three-level designs can have three factors that quantify response and factor relationships. RSM includes MLR, OLS, PCR, and PLS linear designs, non-linear regression analysis, and non-parametric methods such as ACE, NPLS, and MARS.
Scree-plot
Residual variance approaches constancy, and plotted slope levels off, depending on number of components {Scree-test, vision}.
singular-value decomposition
In unsupervised linear methods, correlation matrix can become product of score, eigenvalues, and loading matrices, with diagonalization using QR algorithm.
spectral-mapping analysis
Factor-analysis methods can first take data logarithms to eliminate outliers and then subtract means from rows and columns, to leave only variation, showing which variables are important and how much.
structure space
Spaces can have two or three principal components.
target-transformation factor analysis
Methods can rotate features to match known patterns, such as hypotheses or signatures.
Unsupervised Method
Factors and response variables can relate, without using factor information or predetermined models.
Methods {eight point algorithm} can find structures from motions.
People can recognize six basic facial expressions {facial expression recognition}: anger, disgust, fear, happiness, sadness, and surprise. Expressions have unique muscle activities {Facial Action Coding System}, grouped into Action Parts. Methods detect faces, extract features, and classify expressions. Classifying can use Gabor filters, Bézier volumes, Bayes and Bayesian network classifiers, and Hidden Markov Models.
Gaussian filtering {Kalman filter} can use mean and variance parameters for normal distributions and can increase feature or pixel gain. Kalman filters are parametric, as opposed to particle filters. Kalman filters predict output from input.
First computer-vision stage is to find features, including invariants {local image analysis}. Invariants can be angles, local phase, and orientation.
Distributions can have representations as finite numbers of samples {particle, sample} defined on Markov chains {particle filtering}, rather than using parameters.
Operators {Sobel edge operator} can detect edges.
Methods {support-vector machine} can detect shapes from image segmentation, using color, shape, and distances.
3-Computer Science-Systems-Computer Vision-Algorithms
Outline of Knowledge Database Home Page
Description of Outline of Knowledge Database
Date Modified: 2022.0225