In medical image retrieval, the images are usually subject to a large range of variability. In order to classify medical images, we therefore propose the use of local representations, which are small square windows taken from the images. This approach is combined with a fast approximate k-nearest neighbor technique and yields state-of-the-art results on a medical image database of 1617 images.

Previous work in hypertext classification has resulted in two principal approaches for incorporating information about the graph properties of the Web into the training of a classifier. The first approach uses the complete text of the neighboring pages, whereas the second approach uses only their class labels. In this paper, we argue that both approaches are unsatisfactory: the first one brings...

We present a framework that jointly learns and then uses multiple image windows for improved classification. Apart from using the entire image content as context, class-specific windows are added, as well as windows that target class pairs. The location and extent of the windows are set automatically by handling the window parameters as latent variables. This framework makes the following contr...

In the domain of many relevant classification problems, classes are organized in hierarchies, representing specialization relationships between them. These are the so-called hierarchical classification problems. Methods based on different approaches have been used to solve them, trying to achieve better predictive performance. In this work, we propose two local per level hierarchical classifier...

We present a new method to compute depth-three threshold circuits for pattern classi cation problems. The rst layer of the circuits is calculated from a sample set S of the classi cation problem by a local search strategy that minimises the error on S for each individual gate. The local search is based on simulated annealing with the logarithmic cooling schedule c(k) = = ln (k + 2). The paramet...

The nearest neighbor technique is a simple and appealing method to address classification problems. It relies on the assumption of locally constant class conditional probabilities. This assumption becomes invalid in high dimensions with a finite number of examples due to the curse of dimensionality. Severe bias can be introduced under these conditions when using the nearest neighbor rule. The e...

We introduce a rigorous mathematical theory for the analysis of local histograms, and study how they interact with textures that can be modeled as occlusions of simpler components. We first show how local histograms can be computed as a system of convolutions and discuss some basic local histogram properties. We then introduce a probabilistic, occlusion-based model for textures and formally dem...

Object classification in digital images remains one of the most challenging tasks in computer vision. Advances in the last decade have produced methods to repeatably extract and describe characteristic local features in natural images. In order to apply machine learning techniques in computer vision systems, a representation based on these features is needed. A set of local features is the most...