Principal Component Analysis applied to Real Images

Gerard Brunet (*),  Abdellah Qannari (*)

The purpose of the project is to build a fast and reliable software for recognition
of objects in aerial images. The preliminary stages consist of an edge detection by the
analysis of a color gradient vector. After edge detection and feature extraction of the
three-dimensional color histogram a set of parameters is extracted. The parameters taken
into account to recognize regions are the histograms of the R, G, and B color components
of the pixels and characteristics on the homogeneity and the shape. Principal Component
Analysis was performed for examining relationships among several quantitative variables.
The programmation language used operationally to translate is Java, Software Development
Environment was Eclipse platform with Swing components for the Human Machine Interface.
The software was applied to 400*400 pixels images. Real images have been tested to
evaluate the performance concerning translation, rotation, zoom, and random noise
addition. For reason of speed, the analysis can be made on only a part of the image,
until 1/8 of the image.

(*) University of Poitiers, Niort, France