Classification Stage

Classification Stage
Each pixel is categorised into landcover class to which it closely resembles. If the pixel is not similar to the training data, then it is labeled as unknown. Numerical mathematical approaches to the spectral pattern recognition have been classified into various categories.
Measurements on Scatter Diagram
Each pixel value is plotted on the graph as the scatter diagram indicating the category of the class. In this case the 2-dimensional digital values attributed to each pixel is plottes on the graph


Minimum Distance to Mean Classifier/Centroid Classifier
This is a simple classification strategies. First the mean vector for each category is determined from the average DN in each band for each class. An unknown pixel can then be classified by computing the distance from its spectral position to each of the means and assigning it to the class with the closest mean. One limitation of this technique is that it overlooks the different degrees of variation.


Parallelpiped Classifier
For each class the estimate of the maximum and minimum DN in each band is determine. Then parallelpiped are constructeds o as to enclose the scatter in each theme. Then each pixel is tested to see if it falls inside any of the parallelpiped and has limitation
A pixel may fall outside the parallelpiped and remained unclassified.
Theme data are so strongly corrected such that a pixel vector that plots at some distance from the theme scatter may yet fall within the decision box and be classified erroneously.
Sometimes parallelpiped may overlap in which case the decision becomes more complicated then boundary are slipped.


Gaussian Maximum Likelihood Classifier
This method determines the variance and covariance of each theme providing the probability function. This is then used to classify an unknown pixel by calculating for each class, the probability that it lies in that class. The pixel is then assigned to the most likely class or if its probability value fail to reach any close defined threshold in any of the class, be labeled as unclassified. Reducing data dimensionally before hand is aone approach to speeding the process up.
Unsupervised Classification
This system of classification does not utilize training data as the basis of classification. This classifier involves algorithms that examine the unknown pixels in the image and aggregate them into a number of classes based on the natural groupings or cluster present in the image. The classes that result from this type of classification are spectral classes. Unsupervised classification is the identification, labeling and mapping of these natural classes. This method is usually used when there is less information about the data before classification.

There are several mathematical strategies to represent the clusters of data in spectral space.
Sequential Clustering
In this method the pixels are analysed one at a time pixel by pixel and line by line. The spectral distance between each analysed pixel and previously defined cluster means are calculated. If the distance is greater than some threshold value, the pixel begins a new cluster otherwise it contributes to the nearest existing clusters in which case cluster mean is recalculated. Clusters are merged if too many of them are formed by adjusting the threshold value of the cluster means.


Statistical Clustering
It overlooks the spatial relationship between adjacent pixels. The algorithm uses 3x3 windows in which all pixels have similar vector in space. The process has two steps
Testing for homogeneity within the window of pixels under consideration.
Cluster merging and deletion


Here the windows are moved one at time through the image avoiding the overlap. The mean and standard derivation are calculated for each band of the window. The smaller the standard deviation for a given band the greater the homogenity of the window. These values are then compared by the user specified parameter for delineating the upper and lower limit of the standard deviation. If the window passes the homogenity test it forms cluster. Clusters are created untill then number exceeds the user defined maximum number of clusters at which point some are merged or deleted according to their weighting and spectral distances.


Iso Data Clustering (Iterative Self Organising Data Analysis Techniques)
Its repeatedly performs an entire classification and recalculates the statistics. The procedure begins with a set of arbitrarily defined cluster means, usually located evenly through the spectral space. After each iteration new means are calculated and the process is repeated until there is some difference between iterations. This method produces good result for the data that are not normally distributed and is also not biased by any section of the image.


RGB Clustering
It is quick method for 3 band, 8 bit data. The algorithm plots all pixels in spectral space and then divides this space into 32 x 32 x 32 clusters. A cluster is required to have minimum number of pixels to become a class. RGB Clustering is not baised to any part of the data.