Volume 10 (2020): Issue 2 (April 2020)

In this paper, we look closely at the issue of contaminated data sets, where apart from legitimate (proper) patterns we encounter erroneous patterns. In a typical scenario, the classification of a contaminated data set is always negatively influenced by garbage patterns (referred to as foreign patterns). Ideally, we would like to remove them from the data set entirely. The paper is devoted to comparison and analysis of three different models capable to perform classification of proper patterns with rejection of foreign patterns. It should be stressed that the studied models are constructed using proper patterns only, and no knowledge about the characteristics of foreign patterns is needed. The methods are illustrated with a case study of handwritten digits recognition, but the proposed approach itself is formulated in a general manner. Therefore, it can be applied to different problems. We have distinguished three structures: global, local, and embedded, all capable to eliminate foreign patterns while performing classification of proper patterns at the same time. A comparison of the proposed models shows that the embedded structure provides the best results but at the cost of a relatively high model complexity. The local architecture provides satisfying results and at the same time is relatively simple.

The social learning mechanism used in the Particle Swarm Optimization algorithm allows this method to converge quickly. However, it can lead to catching the swarm in the local optimum. The solution to this issue may be the use of genetic operators whose random nature allows them to leave this point. The degree of use of these operators can be controlled using a neuro-fuzzy system. Previous studies have shown that the form of fuzzy rules should be adapted to the fitness landscape of the problem. This may suggest that in the case of complex optimization problems, the use of different systems at different stages of the algorithm will allow to achieve better results. In this paper, we introduce an auto adaptation mechanism that allows to change the form of fuzzy rules when solving the optimization problem. The proposed mechanism has been tested on benchmark functions widely adapted in the literature. The results verify the effectiveness and efficiency of this solution.

Large-scale image repositories are challenging to perform queries based on the content of the images. The paper proposes a novel, nested-dictionary data structure for indexing image local features. The method transforms image local feature vectors into two-level hashes and builds an index of the content of the images in the database. The algorithm can be used in database management systems. We implemented it with an example image descriptor and deployed in a relational database. We performed the experiments on two image large benchmark datasets.

Nowadays, unprecedented amounts of heterogeneous data collections are stored, processed and transmitted via the Internet. In data analysis one of the most important problems is to verify whether data observed or/and collected in time are genuine and stationary, i.e. the information sources did not change their characteristics. There is a variety of data types: texts, images, audio or video files or streams, metadata descriptions, thereby ordinary numbers. All of them changes in many ways. If the change happens the next question is what is the essence of this change and when and where the change has occurred. The main focus of this paper is detection of change and classification of its type. Many algorithms have been proposed to detect abnormalities and deviations in the data. In this paper we propose a new approach for abrupt changes detection based on the Parzen kernel estimation of the partial derivatives of the multivariate regression functions in presence of probabilistic noise. The proposed change detection algorithm is applied to oneand two-dimensional patterns to detect the abrupt changes.

The paper presented here describes a new practical approach to the reconstruction problem applied to 3D spiral x-ray tomography. The concept we propose is based on a continuous-to-continuous data model, and the reconstruction problem is formulated as a shift invariant system. This original reconstruction method is formulated taking into consideration the statistical properties of signals obtained by the 3D geometry of a CT scanner. It belongs to the class of nutating reconstruction methods and is based on the advanced single slice rebinning (ASSR) methodology. The concept shown here significantly improves the quality of the images obtained after reconstruction and decreases the complexity of the reconstruction problem in comparison with other approaches. Computer simulations have been performed, which prove that the reconstruction algorithm described here does indeed significantly outperforms conventional analytical methods in the quality of the images obtained.