11th WSEAS International Conference on Neural Networks/11th WSEAS International Conference on Evollutionary Computing/11th WSEAS International Conference on Fuzzy Systems, Iasi, Romania, 13 - 15 June 2010, pp.250-255
Given outdoor vehicle images, we try to find an acceptable method chain that maximizes the vehicle color recognition success. Our aim is to determine the color of the vehicle located in a colored image and to make a decision among the chosen seven color classes. At this study, performances of different feature sets obtained by various color spaces and different classification methods are taken to account in order to improve the outdoor vehicle color recognition. Also, different Region of Interest (ROT) and feature vector construction methods are developed for gain better performance. We examined two ROT (smooth hood peace and semi front vehicle), three classification methods (K-Nearest Neighbors, Artificial Neural Networks, and Support Vector Machines), and all possible combinations of sixteen color space components as different feature sets. We obtained 83.50% success in our experiments. As a result, the best performer combination of the classifier, the choice of the ROI, and the feature vector are demonstrated.