TY - GEN
T1 - Textural feature selection for enhanced detection of stationary humans in through-The-wall radar imagery
AU - Chaddad, A.
AU - Ahmad, F.
AU - Amin, M. G.
AU - Sevigny, P.
AU - Difilippo, D.
N1 - Copyright:
Copyright 2014 Elsevier B.V., All rights reserved.
PY - 2014
Y1 - 2014
N2 - Feature-based methods have been recently considered in the literature for detection of stationary human targets in through-The-wall radar imagery. Specifically, textural features, such as contrast, correlation, energy, entropy, and homogeneity, have been extracted from gray-level co-occurrence matrices (GLCMs) to aid in discriminating the true targets from multipath ghosts and clutter that closely mimic the target in size and intensity. In this paper, we address the task of feature selection to identify the relevant subset of features in the GLCM domain, while discarding those that are either redundant or confusing, thereby improving the performance of feature-based scheme to distinguish between targets and ghosts/clutter. We apply a Decision Tree algorithm to find the optimal combination of co-occurrence based textural features for the problem at hand. We employ a K-Nearest Neighbor classifier to evaluate the performance of the optimal textural feature based scheme in terms of its target and ghost/clutter discrimination capability and use real-data collected with the vehicle-borne multi-channel through-The-wall radar imaging system by Defence Research and Development Canada. For the specific data analyzed, it is shown that the identified dominant features yield a higher classification accuracy, with lower number of false alarms and missed detections, compared to the full GLCM based feature set.
AB - Feature-based methods have been recently considered in the literature for detection of stationary human targets in through-The-wall radar imagery. Specifically, textural features, such as contrast, correlation, energy, entropy, and homogeneity, have been extracted from gray-level co-occurrence matrices (GLCMs) to aid in discriminating the true targets from multipath ghosts and clutter that closely mimic the target in size and intensity. In this paper, we address the task of feature selection to identify the relevant subset of features in the GLCM domain, while discarding those that are either redundant or confusing, thereby improving the performance of feature-based scheme to distinguish between targets and ghosts/clutter. We apply a Decision Tree algorithm to find the optimal combination of co-occurrence based textural features for the problem at hand. We employ a K-Nearest Neighbor classifier to evaluate the performance of the optimal textural feature based scheme in terms of its target and ghost/clutter discrimination capability and use real-data collected with the vehicle-borne multi-channel through-The-wall radar imaging system by Defence Research and Development Canada. For the specific data analyzed, it is shown that the identified dominant features yield a higher classification accuracy, with lower number of false alarms and missed detections, compared to the full GLCM based feature set.
KW - Through-The-wall radar imaging
KW - co-occurrence matrix
KW - feature selection
KW - target detection
UR - http://www.scopus.com/inward/record.url?scp=84905706803&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84905706803&partnerID=8YFLogxK
U2 - 10.1117/12.2049416
DO - 10.1117/12.2049416
M3 - Conference contribution
AN - SCOPUS:84905706803
SN - 9781628410143
T3 - Proceedings of SPIE - The International Society for Optical Engineering
BT - Radar Sensor Technology XVIII
PB - SPIE
T2 - Radar Sensor Technology XVIII
Y2 - 5 May 2014 through 7 May 2014
ER -