In this paper, ensembles of k-nearest neighbors classifiers are explored for gene expression cancer classification, where each classifier is linked to a randomly selected subset of genes. It is experimentally demonstrated using five datasets that such ensembles can yield both good accuracy and dimensionality reduction. If a characteristic called dataset complexity guides which random subset to include into an ensemble, then the ensemble achieves even better performance.

Gene expression based cancer classification using classifier ensembles is the main focus of this work. A new ensemble method is proposed that combines predictions of a small number of k-nearest neighbor (k-NN) classifiers with majority vote. Diversity of predictions is guaranteed by assigning a separate feature subset, randomly sampled from the original set of features, to each classifier. Accuracy of k-NNs is ensured by the statistically confirmed dependence between dataset complexity, determining how difficult is a dataset for classification, and classification error. Experiments carried out on three gene expression datasets containing different types of cancer show that our ensemble method is superior to 1) a single best classifier in the ensemble, 2) the nearest shrunken centroids method originally proposed for gene expression data, and 3) the traditional ensemble construction scheme that does not take into account dataset complexity.

Detection of text from documents in which text is embedded in complex colored and textured backgrounds is a very challenging problem. In this paper, we propose a simple texture-based approach based on edge information for this task. The performance of our method is compared to that obtained by a method based on the discrete cosine transform which was recently proposed by Y. Zhong et al. (2000) for text localization in compressed digital video. In our experiments, both methods performed about equally well for small-sized text, but our method was better in the case of large-sized text. The principal advantage of our approach is that in addition to the text detection problem, the same edge representation can also be used for other image interpretation tasks.

Generation of ground-truths is of great importance for unbiased performance evaluation of document layout analysis methods. This is especially necessary because many methods are claimed to be skew-tolerant. However, experimental evaluation of this fact is often based only on human subjective judgement and restricted to a few experiments. The main obstacle for obtaining human-independent and more automated performance evaluation is that usually there are only ground-truths for upright images, i.e., images with no skew of text lines, because currently available ground-truthing techniques are too time-consuming. We propose a methodology of automatic generation of ground-truths for skewed images by using the ground-truths available for upright images. This methodology is simple and quite fast because processing is done at the level of small square blocks, but not at pixel level.

Presents a new technique for estimating an arbitrary skew angle using a simple and efficient text row accumulation based on first- and second-order statistics. In contrast to other works, where an image resolution of 100-300 dpi is necessary, a lower resolution (50 dpi) is enough for our method to obtain a correct result. The performance of the method was evaluated on the database UW-1 of the University of Washington and compared with an advanced Hough transform-based method. A significantly lower error rate for our method confirms its advantages and benefits for automated document processing.

The authors describe a system for automatic interpretation of map-drawings. To vectorize large-size 2-D line-drawing images, an effective pipeline-oriented scheme and special techniques have been developed on the basis of economic image representation. This allows processing large-size binary images on a personal computer and obtaining satisfactory time characteristics. This approach can be easily realized in a pipelined architecture using a simple transputer chain. The document object recognition techniques are described.>