ROLE OF TEXTURE IN COMPUTER AIDED SIGNATURE ANALYSIS - 
A CASE STUDY 
by 
R.K.Aggarwala, Survey of India, Dehra Dun, India 
Abstract: It is well known that the texture plays a significant 
role in identification and classification of land uses/crop types 
using manual interpretation techniques. In computer aided automated 
interpretation methods, however, signature analysis by multispectral 
reflectance tonal data only has been extensively used. Very few 
studies exist where texture has been quantized and successfully 
used in automated discrimination and classification procedures. 
The prime difficulty apparently has been the selection of suitable 
numerical analogue (s) of texture which are convenient both of 
definition and measurement without undue complexity. 
In this study, "standard deviation of photographic density within 
a single field" has been used as a measure of texture in signature 
analysis of five crop types - uncut corn, cut corn, alfalfa, wheat 
field and idle fields. The study was directed to investigate the 
role of this measure of texture in improving discrimination and 
classification accuracy for these crops as a case study. "Standard 
deviation" as a measure of texture was selected for its inherent 
simplicity and ease of determination without additional measurements. 
It is expected that the results of the findings could be extended 
to other "area targets" such as used for broad land-use classifi- 
cation, or for mapping other crop types. 
  
  
A single aerial photographic negative flown in September, at scale 
of 1:14.700 was selected for the study. Sensitometrically controlled 
photographic prints at three scales (1:14,000, 1:7350, 1:3675) from 
the same negative and a fixed densitometric aperture size of 1.5 mm 
were used for densitometric tonal reflectance measurements. 
Thirteen replicates per crop type were sampled. Well distributed 
tonal measurements were made on each field providing thirteen sets 
of mean tonal and textural values at each scale. Statistical analysis 
(Analysis of variance, discriminant analysis and discriminant scores, 
correlation analysis, canonical analysis and scores) were carried 
out, using the univariate/multivariate data. Bivariate plots of 
tonal/textural variables and of the first two canonical variables 
were drawn using the computer. 
It was found that within the fixed design of the experimental study, 
both tone and texture were about equally effective in discriminating 
all the crop types simultaneously. Pairwise discriminant analysis 
indicated that the role of texture was largely complementary to 
that of tone for certain crop types. In combination (tone * texture) 
the classification results improved by about 50$ at each scale. The 
results were better for larger scales. Combination of variable 
values at more than one scale improved the accuracy by almost 100$. 
From the results and analysis, it is concluded that texture as 
defined by the "standard deviation of tone" has a significant 
capability to improve crop or land use identification. It is also 
noted that change of scale (even from the same negative) provided 
additional un-correlated information. This study thus provides a