You are on page 1of 9

NIAGARA COLLEGE GIS-GM

Digital Image
Processing
Principal Component Analysis
Assignment #2
By: Marc Michael Mancino
For: Janet Finlay

10 Jessica Drive St. Catharines Ontario L2M 6V7


(905) 935-8366 marcmancino@gmail.com
February 16th, 2014
GISC9216D2
Janet Finlay
Niagara College
135 Taylor Road
Niagara-on-the-Lake
Dear Janet Finlay:

RE: GISC9216D1 Digital Image Processing

The purpose of this deliverable is to perform a principal component analysis


on the subset raster image, classify it and then compare the obtained
results. The report includes all the answers to all the questions noted in the
terms of reference. The process for conducting a principal component
analysis has been proven useful in solidifying my knowledge of ERDAS
Imagine and digital image interpretation and further enhances my ability to
perform like a GIS expert.
Should you require further information, please do not hesitate to contact me
at your convenience. I look forward to your comments and suggestions on
this deliverable.
Sincerely,
Marc Mancino
MM/
Enclosures:1.) Answers to the questions.

Questions:
1) Why do we need to transform original image bands to the Principal
Components?
We need to transform the original image bands to the Principal
Components so the redundancies between the correlations of the
bands are discarded and produce an image with the potential for
more easily distinguishable features.

2) After exploring the original image, it was determined that the bands
with a strong correlation are: (1, 2), (1, 3), and (2, 3), as shown below
in Table 1.
Table 1 - Resultant Scatter graphs of the Strongly Correlated Bands for the
Subset Raster Image

(1, 2)

Strongly Correlated Bands


(1, 3)

(2, 3)

3) After performing the PCA transformation, discuss the variance shown


on the first, second and third channel of the PCA result.
The first channel of the PCA result contains the majority of the
variance at 88.55%. The second channel of the PCA result contains
10.04% of the variance. The third channel of the PCA result contains
1.10% of the variance. Altogether, the first three channels of the PCA
result account for 99.69% of the variance as indicated below in Table
2.

Table 2 Resultant Eigenvalues and Respective % Variance for the Principal


Component Analysis

1
2
3
4
5
6
Total

Eigenvalue
3499.008686
396.9109978
43.42106486
6.73981675
3.523588001
1.844049527
3951.448203

% Variance
88.550033
10.044697
1.09886458

99.6935945

4) Compare the original data to the PCA channels.


As previously mentioned, the strongly correlated bands for the
subset raster image were (1, 2), (1, 3), and (2, 3). After performing
the Principal Component Analysis and producing a feature space
image, shown below in Table 3, it is evident that the correlation
between the same three bands is no longer existent.
Table 3 Resultant Scatter graphs for the Strongly Correlated Bands after
the Principal Component Analysis

(1, 2)

Strongly Correlated Bands Post-PCA


(1, 3)

(2, 3)

5) An unsupervised classification was performed on the original image


and then on the PCA results, keeping all of the classification
parameters the same.

6) Compare the 2 obtained classification results. Discuss how the PCA


process helped (if any) in the classification of Urban vs Agriculture in
the image.
The results of both the original unsupervised classification and PCA
unsupervised classification methods are displayed below in Table 4. A closer
look at the original unsupervised classification, located in the Appendix, is
displayed in Figure 1. A closer look at the PCA unsupervised classification,
located in the Appendix, is displayed in Figure 2.
Table 4 - Resultant Images for the Original Unsupervised and PCA
Unsupervised Classifications

Original Unsupervised
Classification

PCA Unsupervised Classification

There are very slight differences in the number of pixels between the
signatures of each classification, most of which only deviate by an almost
negligible amount. The black circle indicator in Table 4 shows a slight
difference in pixel density for the shallow coast (cyan) feature. The PCA
seemed to put less emphasis on pixel density in that area. Any other pixel
differences are negligible, but are evident in the attribute tables of each
image, as indicated below in Table 5. The original unsupervised
classification looks nearly identical, which could indicate that the
procedures for conducting it were followed diligently and in turn produced
a highly refined result.

Table 5 - Attribute Tables for each Classification Displaying Pixel


Distribution for each Signature

Original Unsupervised Classification

PCA Unsupervised Classification

Appendix:

Figure 1 Original Unsupervised Classification of the AOI Using


ERDAS Imagine 2014

Figure 2 - Unsupervised Classification of the Principal Components


for the AOI Using ERDAS Imagine 2014

You might also like