Title :
A Context Classifier
Author :
Haralick, Robert M. ; Joo, Hyonam
Author_Institution :
Department of Electrical Engineering, University of Washington, Seattle, WA 98195
Abstract :
All other things being equal, context classifiers have a higher identification accuracy than pixel independent classifiers. This paper discusses a classifier whose context for each pixel is the best row monotonically increasing path including the given pixel. Use of such a context results in a two pass algorithm in which both the top down and bottom up pass require only two rows of data and whose computational complexity per pixel is constant independent of the size of image or the length of the best context path including the given pixel.
Keywords :
Computational complexity; Decision making; Frequency; Labeling; Layout; Machine vision; Pattern recognition; Pixel; Stochastic processes; Text recognition;
Journal_Title :
Geoscience and Remote Sensing, IEEE Transactions on
DOI :
10.1109/TGRS.1986.289563