Over the past two decades, image retrieval from large databases has become a popular application of machine learning, especially when applied to web-based databases. Wavelet transformation is one of the more recent tools in image processing, and is mainly used as a platform for extracting features. The quality of image retrieval algorithms depends heavily on the quality of the features used. Computation time for image retrieval is reduced with fewer features, while the retrieval rate is improved with an increased ability to characterize the features.

In this paper, the authors identify two sets of shape-based features: local and global. Global features are extracted with grid-based methods and are made invariant in size, translation, and rotation. Local features are extracted with wavelet transformation and singular value decomposition. Experimental results are presented in terms of precision using similarity measurements.

The authors touch on almost all phases of a typical image retrieval system, starting from preprocessing, so this paper does not provide much new or in-depth information. It might have been more interesting if the authors had used other tools of merit, such as receiver operating characteristic (ROC) analysis, to demonstrate the effectiveness of this approach.

Nevertheless, the presentation is clear and well organized. The paper could be used for mini-projects by senior undergraduate and graduate students.