Search

This paper presents paper retrieval using the specific paper features chain and laid lines. Paper features are detected in digitized paper images and they are represented such that they could be used for retrieval. Optimal retrieval performance is achieved by means of a trainable similarity measure for a given set of paper features. By means of these methods a retrieval system is developed that art experts could use real-time in order to speed up their paper research. [ABSTRACT FROM AUTHOR]

Co-occurrence matrices are proved to be useful tool for the purpose of texture recognition. However, they are sensitive to the change of the illumination conditions. There are standard preprocessing approaches to this problem. However, they are lacking certain qualities. We studied the tangent kernel SVM approach as an alternative way of building illumination-robust texture classifier. Testing on the standard texture data has shown promising results. [ABSTRACT FROM AUTHOR]

An industrial rock classification system is constructed and studied. The local texture information in many image patches is extracted and classified. The decisions made at the local level are fused to form the high-level decision on the image/rock as a whole. The main difficulties of this application lay in significant variability and inhomogeneity of local textures caused by uneven rock surfaces and intrusions. Therefore, an emphasis is paid to the derivation of informative representation of local texture and to robust classification algorithms. The study focuses on the co-occurrence representation of texture comparing the two frequently used strategies, namely the approach based on Haralick features and methods utilizing directly the co-occurrence likelihoods. Apart of maximum-likelihood (ML) classifiers also an alternative method is studied considering the likelihoods to prototypes as feature of a new space. Unlike the ML methods, a classifier built in this space may leverage all training examples. It is experimentally illustrated, that in the rock classification setup the methods directly using the co-occurrence estimates outperform the feature-based techniques. [ABSTRACT FROM AUTHOR]

A typical recognition system consists of a sequential combination of two experts, called a detector and classifier respectively. The two stages are usually designed independently, but we show that this may be suboptimal due to interaction between the stages. In this paper we consider the two stages holistically, as components of a multiple classifier system. This allows for an optimal design that accounts for such interaction. An ROC-based analysis is developed that facilitates the study of the inter-stage interaction, and an analytic example is then used to compare independently designing each stage to a holistically optimised system, based on cost. The benefit of the proposed analysis is demonstrated practically via a number of experiments. The extension to any number of classes is discussed, highlighting the computational challenges, as well as its application in an imprecise environment. [ABSTRACT FROM AUTHOR]

Unlike fixed combining rules, the trainable combiner is applicable to ensembles of diverse base classifier architectures with incomparable outputs. The trainable combiner, however, requires the additional step of deriving a second-stage training dataset from the base classifier outputs. Although several strategies have been devised, it is thus far unclear which is superior for a given situation. In this paper we investigate three principal training techniques, namely the re-use of the training dataset for both stages, an independent validation set, and the stacked generalization. On experiments with several datasets we have observed that the stacked generalization outperforms the other techniques in most situations, with the exception of very small sample sizes, in which the re-using strategy behaves better. We illustrate that the stacked generalization introduces additional noise to the second-stage training dataset, and should therefore be bundled with simple combiners that are insensitive to the noise. We propose an extension of the stacked generalization approach which significantly improves the combiner robustness. [ABSTRACT FROM AUTHOR]