ERSC 3P07 Lecture Notes - Lecture 7: Multispectral Image, Euclidean Distance, Scatter Plot
19 May 2016
School
Department
Course
Professor
Document Summary
This classifier assigns each unknown pixel (e. g. , pixel a) to the closest class mean. Limitations: this classifier is insensitive to class variability (i. e. , it does not take class variability into account) Therefore, pixel a would be assigned to class although it probably belongs to class . This classifier takes class variance into account by considering the range of brightness values for each class training set. Range is defined as the highest and lowest brightness values in each band. These typically produce rectangular regions in multi-dimensional feature space called parallelepipeds. Using this classifier, unknown pixels are classified based on the parallelepiped (decision region or boundary) into which it falls. Limitations: overlapping pixels and pixels lying outside of the decision boundaries are labeled as. Here, pixel b would be assigned to class even though it likely belongs to class . Unknown pixels that occur in areas of overlap are either labeled as one or both class or it is labeled unclassified.
