Search

Vehicle type/make recognition based on images captured by surveillance cameras is a challenging task in intelligent transportation system and automatic surveillance. In this paper, we comparatively studied two feature extraction methods for image description, i.e. a new multiresolution analysis tool called Fast Discrete Curvelet Transform and the pyramid histogram of oriented gradients (PHOG). Curvelet Transform has better directional and edge representation abilities than widely used wavelet transform, which is particularly appropriate for the description of images rich with edges. PHOG represents the local shape by a histogram of edge orientations computed for each image sub-region, quantized into a number of bins, thus has the ascendency in its description of more discriminating information. A composite feature description from PHOG and Curvelet can further increase the accuracy of classification by taking their complementary information. We also investigated the applicability of the Rotation Forest (RF) ensemble method for vehicle classification based on the combined features. The RF ensemble contains a set of base multilayer perceptrons which are trained using principal component analysis to rotate the original axes of combined features of vehicle images. The class label is assigned by the ensemble via majority voting. Experimental results using more than 600 images from 21 makes of cars/vans show the effectiveness of the proposed approach. The composite feature is better than any single feature in the classification accuracy and the ensemble model produces better performance compared to any of the individual neural network base classifier. With a moderate ensemble size of 20, the Rotation Forest ensembles offers a classification rate close to 96.5%, exhibiting promising potentials for real-life applications. [ABSTRACT FROM AUTHOR]

In this paper, a hybrid feature extraction technique using 2D principal component analysis (2DPCA) and discrete orthogonal Krawtchouk moment (KM) are used to extract the global and local features from the face. Ensemble of RBF classifiers are used to classify the image. Decision-level fusion is done using fuzzy integral to generate more accurate classification than each of the constituent classifiers. The proposed system is evaluated using ORL and YALE databases. Experimental results show that the combination of global and local features promotes the system performance. The fusion of multiple RBFs using fuzzy integral performed better as compared to conventional aggregation rules. [ABSTRACT FROM AUTHOR]