Ilya Levner, Vadim Bulitko
{ilya | bulitko}@cs.ualberta.ca
University of Alberta
Department of Computing Science
Edmonton, Alberta, T6G 2E8, CANADA

Automated image interpretation is an important task in numerous applications ranging from security systems to natural resource inventory based on remote-sensing. Recently, a second generation of adaptive machine-learned image interpretation systems have shown promising performance in several challenging domains. While demonstrating an unprecedented improvement over hand-engineered and first generation machine-learned systems in terms of cross-domain portability, design-cycle time, and robustness, such systems are still severely limited. This paper reviews the anatomy of the Multi Resolution Adaptive Object Recognition framework (MR ADORE) and presents research aimed at removing the last vestiges of human intervention [1]. More specifically, feature selection/extraction is performed manually by human domain experts thereby hindering automated creation of image interpretation systems [2]. The goal of our research is to develop techniques that can automatically select/extract relevant features solely from annotated image pairs used to train the object recognition system . To date, principal component analysis (PCA) [3] has been evaluated against hand-crafted features, as well as raw pixels (i.e., no feature extraction) [4]. The experimental results indicate that naive usage of PCA as a feature creation technique presents several challenges. First, and foremost, the linear manifold created out of non-registered aerial photographs does not capture enough relevant information. As a result non-linear manifold learning methods, such as kPCA, Isomap and Local Linear Embedding, need to be applied in order to approximate training data with sufficient accuracy. In summary, this paper focuses on autonomous feature extraction methods aimed at removing the need
for human expertise in the feature selection process.

Keywords: Feature extraction, Dimensionality reduction, AI approaches to computer vision, Object recognition.

References

[1] V. Bulitko, G. Lee, I. Levner, L. Li, and R. Greiner. Adaptive image interpretation: A spectrum of machine learning problems. In Proceedings of International Conference on Machine Learning, Workshop on The Continuum from Labeled to Unlabeled Data in Machine Learning and Data Mining, Washington, D.C., 2003.

[2] B. A. Draper. From knowledge bases to Markov models to PCA. In Proceedings of Workshop on Computer Vision System Control Architectures, Graz, Austria, 2003.

[3] M. Kirby. Geometric Data Analysis: An Emprical Approach to Dimensionality Reduction and the Study of Patterns. John Wiley & Sons, New York, 2001.

[4] I. Levner, V. Bulitko, G. Lee, L. Li, and R. Greiner. Automated feature extraction for object recognition. In Proceedings of the Image and Vision Computing New Zealand conference, Palmerson North, NZ, 2003.