Digital image clasification is very useful for extraction of information from images. In this process, the human element plays a crucial role, although there are instances that computers may be used for visualization and digitalization, the interpretation (for any outcomes) is done mannually. This makes human operator key in this process, as the operator instructs the computer to perform an interpretation according to certain conditions defined by the operator.
This learning path will introduce you to the process, principles and techniques of image classification. In this case, we will first look at what Digital image classification is all about in this path and subsequently consider other elements associated with it.
In this learning path another element worth of consumption is the term "Image space", which is defined by the spatial distribution of Digital Numbers (DNs) or the values of a pixel.
It is crucial to consider "training samples" in this learning path, which has to do with the elements use to define clusters in the feature space.
Again the term "Feature space" , which represents the graph that shows the feature vectors, is the next item in this learning path worth of consideration. Feature space is also known as Feature space plot or Scatter plot.
Supervised Image Classification is usually done where the operator defines the spectral characteristics of the classes by identifying sample areas. This approach of image classification requires that the operator is familiar with the area of interest.
This then brings us to Unsupervised Classification, where clustering algorithms are used to partition the feature space into a number of clusters. When there is insufficient knowledge available about the area of interest or the classes of interest have not yet been defined.
The next concept that this learning path will introduce you to is called "Classification algorithm", which is used as an application for the classification of an image. The choice of the algorithm depends on the purpose of the classification, the characteristics of the image and training data. Basically the three (3) classifier algorithms are: *Box classifier; *Minimum distance to mean; and *Maximum likelihood classifiers.
As image classification is based on samples of the classes, the actual quality of the classification results should be checked, hence the need for us to consider "Validation of the results" as part of this learning path. In the validation of the result process, Reference set (ground truth data) is used to assess the quality of an image being classified.
In this learning path, the next subject matter for our discussion is the Pixel based classification which helps us as a technique to derive thematic classes from multi-band images. In its application it is proned to certain limitations, namely --- it results in *spectral classes, and that * each pixel is assigned to one class only. Hence alternative technique of classification is Object based classification, which we will consider as the next concept in this path.
Object based classification is the last concept for our consideration in this learning path, which breaks down an image into spectrally homogeneous segments that correspond to fields - examples buildings, tree strands etc. It is also called Segmentation-based Analysis.