Eric Nowak

Randomized Clustering Forests for Image Classification

Some of the most effective recent methods for content-based image classification work by quantizing image descriptors, and accumulating histograms of the resulting visual word codes. Large numbers of descriptors and large codebooks are required for good results and this becomes slow using k-means. We introduce Extremely Randomized Clustering Forests-ensembles of randomly created clustering trees-and show that they provide more accurate results, much faster training and testing, and good resistance to background clutter. Second, an efficient image classification method is proposed. It combines ERC-Forests and saliency maps very closely with the extraction of image information. For a given image, a classifier builds a saliency map online and uses it to classify the image. We show in several state-of-the-art image classification tasks that this method can speed up the classification process enormously. Finally, we show that the proposed ERC-Forests can also be used very successfully for learning distance between images. The distance computation algorithm consists of learning the characteristic differences between local descriptors sampled from pairs of same or different objects. These differences are vector quantized by ERC-Forests and the similarity measure is computed from this quantization. The similarity measure has been evaluated on four very different datasets and always outperforms the state-of-the-art competitive approaches.

Vehicle Categorization: Parts for Speed and Accuracy

In this paper we propose a framework for categorization of different types of vehicles. The difficulty comes from the high inter-class similarity and the high intra-class variability. We address this problem using a part-based recognition system. We particularly focus on the trade-off between the number of parts included in the vehicle models and the recognition rate, i.e the trade-off between fast computation and high accuracy. We propose a high-level data transformation algorithm and a feature selection scheme adapted to hierarchical SVM classifiers to improve the performance of part-based vehicle models. We have tested the proposed framework on real data acquired by infrared surveillance cameras, and on visible images too. On the infrared dataset, with the same speedup factor of 100, our accuracy is 12% better than the standard one-versus-one SVM.