Image classification.
Supervised classification
Unsupervised classification.
Stages:
Raw data
Preprocessing
Signature collection
Signature evaluation
Classification
.
Information Class and Spectral Class
Information class: It is a class specified by an image analyst. It refers to the information to be extracted.
Spectral class: It a class which includes similar gray-level
vectors in the multispectral space. Spectral classes are groups of pixels that are uniform (or near-similar) with respect to their brightness values.
.
Supervised and Unsupervised
Supervised (Information Class)
Have a set of desired dasses in mind then create the appropriate signatures from the data.
Appropriate
• when one wants to identify relatively few dasses
. when one has selected training sites that can be verified with ground truth data
• when one can identify distinct, homogeneous regions that represent each dass.
Unsupervised (Spectral Class)
Classes to be determined by spectral distinctions that are inherent in the data → define the dasses later.
Appropriate-
when one wants to define many dasses easily, and then identify dasses.
.
Training for Classification
Computer system must be trained to recognize patterns in image data.
Process of defining the criteria by which these patterns are recognized.
Supervised Training is controlled by the analyst.
Select pixels that represent patterns instruct the computer system to identify pixels with similar characteristics.
More accurate but requires high skill.
Unsupervised Training is computer-automated.
Specify number of classes the computer uncovers statistical classes.
Less accurate and less skill required.
.
Supervised Classification
Common decision rules:
Parametric decision rules:
Minimum distance classifier / Centroid classifier.
Maximum likelihood / Bayesian classification.
Nonparametric decision rule:
Parallelepiped classifier.
Feature space classifier.
.
Minimum Distance/Centroid Classifier:
Calculates the spectral distance between the candidate pixel and the mean of each signature.
The candidate pixel is assigned to the class with the closest mean.
Calculates mean of the spectral values for the training set in each band and for each category.
Measures the distance from a pixel of unknown identify to the mean of each category.
Assigns the pixel to the category with the shortest distance.
Assigns a pixel as "unknown" if the pixel is beyond the distances defined by the analyst.
Methods of calculation Minimum Spectral Distance:
Euclidean distance: based on the Pythagorean theorem
Mahalanobis distance:
■ Variance-Covariance matrix are used (normal distribution of DN is assumed)
.
Maximum Likelihood Classifier:
• This classifier quantitatively evaluates both the variance and covariance of the trained spectral response patterns when deciding the fate of an unknown pixel.
■To do this the classifier assumes that the distribution of points for each cover-type are normally distributed.
• Under this assumption, the distribution of a category response can be completely described by the mean vector and the covariance matrix.
■ Given these values, the classifier computes the probability that unknown pixels will belong to a particular category.
Probability function is calculated from the inputs.
Assumes probabilities are equal for all dasses.
Each pixel is then judged as to the dass to which it most probably belong.
.
Parallelepiped Classifier:
Based on Maximum and Minimum values in each signature.
..
Feature Space Classifier:
Based on discrete objects (polygons) in a feature space image.
More accurate than parallelepiped.
Comments
Post a Comment