Font Size: 

Knowledge-based Recognition of Objects in Aerial Images

Abdellah Qannari (*),  Gerard Brunet (*),  Mohamed Ibazizen (*)

The project consists in determining regions in aerial images with use of statistical
methods. The statistical methods are the Principal Component Analysis and the  HAC
(Hierarchical Ascending Classification). For this purpose, the preliminary stages consist
of a fast and reliable edge detection, and a construction of regions with analysis of
the three-dimensional color histogram. Principal Component Analysis was performed for
examining relationships among several quantitative variables. The result of the edge
pixel detection procedure is used for clustering-based segmentation in order to extract
meaningful regions representing the objects in the scene. SAS (Statistical Analysis
System) was used extensively and adaptations and specializations of some procedures were
made with SAS/IML package. The software was applied to 400*400 pixels images .The
performance has been evaluated with real images  concerning translation, rotation, zoom,
and random noise addition. A practical operationnal system has been implemented, with
Java language.

(*) University of Poitiers, Niort, France



References for the Project

Hastie, T., Tibshirani, R., Friedman, J., 2001
The Elements of Statistical Learning. Springer

Jaba, E., Qannari, A., 2013.
Analyse Discriminante avec applications sous SPSS et SAS .  Eyrolles, France.
Translated for Editura Economica, Romania.

Busin, L., Vandenbroucke, N., Macaire, L., 2007. Color spaces and image segmentation.
Internal report, UMR CNRS 8146, France.



Software Development Environment

SAS  (Statistical Analysis System)  development environment   9.3
SAS IML  Package  (interactive  Matrix  Language)
Eclipse  Platform development environment
Swing Components graphical user interface
Matlab (interactive Matrix Laboratory), for preliminary steps

classification; real images; statistical methods