8UnsupervisedClassificationLectureNotes

8UnsupervisedClassificationLectureNotes - 1 GISc 4037...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
1 GISc 4037 Digital Image Analysis Dr. Charles Roberts Unit 8 - Unsupervised Classification Introduction to Digital Image Classification Up until the lecture on image ratioing, we have looked at ways of qualitatively classifying the imagery. We have imposed structure on the data, rather than finding structure within the data. Density slicing, contrast enhancement, filtering are techniques we use to impose structure based on our experiences as image analysts. These were informed, but subjective methods of image classification. Most ways of classifying maps are subjective, in that the user decides based on arbitrary criteria, what the classes will be. In digital image analysis, image classification has a special meaning that uses the spectral qualities of the image to classify each pixel individually. These techniques are less subjective than others because the reflectance characteristics of the surface play a major role in the assignment of pixels to land-cover classes, whereas the human role is minimized. So, in Digital Image Analysis, image classification has a special meaning. There are two kinds of image classification techniques, and several varieties of each. These are called Supervised and Unsupervised classification. Today we will discuss Unsupervised Classification. This is the most objective classification technique used in image analysis or for that matter, in Geography, Cartography and G.I.S. today. When we talk about image classification in this class, we are talking about spectral or point classifiers that classify imagery according to local spectral information. We can call these classes Information Classes, which represent recognizable Land Covers from which we infer Land Use. Notice that it is a three step inference. The computer finds information classes, and then the humans name the classes, based on our understanding of: Information Classes:Spectral Characteristics of Land-Covers Land-Use as inferred by Land-Covers Image Classification Algorithms
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
2 Information Classes and Land Cover Classes The role of the image analyst is to determine how many classes there should be, what the variation in the spectral signatures of each class should be, and finally to label classes that the cluster analysis has found in the spectral characteristics of the multivariate data set. It is not unusual to find that a single land cover is represented by more than one spectral class. Take for example a golf course. The area around the hole, the link, the sand traps and the ponds may represent separate spectral classes but only one land-use. Often, we must treat subclasses as classes, then we do density slicing or a piecewise contrast stretch to combine the various subclasses into one called “Golf Courses”. Unsupervised Classification Vs. Supervised Classification
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 12/21/2010 for the course GIS 4037c taught by Professor Roberts during the Fall '10 term at FAU.

Page1 / 12

8UnsupervisedClassificationLectureNotes - 1 GISc 4037...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online