Abstract
In this paper the methods of the extraction of plants cover in an urban area by using an IR colour aerial photograph are discussed. Three colour densities R, G and B are measured and analysed by discriminant analysis to classify into two clusters, plants and other objects.
Some trials transforming the measured three colour densities to other proper quantities are made to reduce the number of variates in discriminant analysis and/or to abate the error. Because the histograms of densities R, G and B do not show a normal distribution, a good result can not be obtained when we use R, G and B as a multivariate vector.
The main results as as follows:
(1) The most suitable coordinates system for separating plants from other objects is logarithmic chromaticity coordinates (B/ (B+G+R), G/ (B+G+R) ), but objects in the shadow and water are misjudged and classified into the plants.
(2) Using the coordinates system (B-R, G-R), corresponding to the Bi-Band ratio, two clusters show a good separation, and moreover (G-R) has no correlation with (B-R) in regard to separation. This fact shows a validity of plants extraction by photographic processing.
(3) An improvement of linear discrminant function by use of the optimal variance-covariance matrix has no notable effect on a decrease in the discriminant error.
