Michal Pryczek

Active contours as knowledge discovery methods

In the paper we show that active contour methods can be interpreted as knowledge discovery methods. Application area is not restricted only to image segmentation, but it covers also classification of any other objects, even objects of higher granulation. Additional power of the presented method is that expert knowledge of almost any type can be used to classifier construction, which is not always possible in case of classic techniques. Moreover, the method introduced by the authors, earlier used only for supervised classification, is here applied in an unsupervised case (clustering) and examined on examples.

Cognitive hierarchical active partitions in distributed analysis of medical images

Semantic oriented image analysis has always been considered a challenging task, as it does not concentrate on segmentation process itself, but on interpretation of various image fragments. Contextuality of the process has recently gained significant research interest, with knowledge from image domain being repeatedly highlighted as crucial in achieving satisfactory method effectiveness. The present article elaborates on the recently described contextual hierarchical active partitions (CHAP) technique and its distributed reformulation. CHAP framework lets domain knowledge to be injected to the automated medical study analysis in a seamless and transparent manner by enabling a human expert to interactively participate in the process, e.g. by solving subtasks currently too difficult to solve by automated agents. Separation of agents makes it easy to design complex analysis algorithms from well tested and predictable components making it easy to inject human expertise at any point as needed.

Supervised web document classification using discrete transforms, active hypercontours and expert knowledge

In this paper, a new method of supervised classification of documents is proposed. It utilizes discrete trasforms to extract features from classified objects and adopts adaptive potential active hypercontours (APAH) for document classification. The idea of APAH generalizes classic contour methods of image segmentation. It has two main advantages: it can use almost any knowledge during the search for an optimal classification function and it can operate in a feature space where only metric is defined. Here, both of them are utilized - the first one by using expert knowledge about significance of documents from training set and the second one by inducing new metrics in feature spaces. The method has been evaluated on the subset of open directory project (ODP) database and compared with k-NN, the well known classification technique.

Cognitive hierarchical active partitions using patch approach

Rapidly developing Medical Image Understanding [1] field requires fast and accurate context and semantic oriented object recognition methods. This is crucial due to both descriptions substantialness requirements and diagnostic responsibility. Synthetic approach to medical image analysis is expected, integrating various kinds of knowledge, to facilitate processing and make the results more meaningful. Cognitive Hierarchical Active Partitions is a flexible image analysis tool, facilitating use of semantic and contextual knowledge encoded in patch based linguistic description of a given image. In the paper presented, this technique is evaluated in ventricular system recognition task on example set of brain CT scans.

On Textual Documents Classification Using Fourier Domain Scoring

Recently, Fourier and cosine discrete transformations have been proposed for textual document ranking. The advantage of the methods is that not only the count of a frequency term within documents used; the spatial information about presence of the term is also considered. Here, this novel approach is used to improve performance of classifiers

Neuronal groups and interrelations

Recent development of various domains of artificial intelligence including information retrieval and text/image understanding created demand on new, sophisticated, contextual methods for data analysis. This article formulates neuronal group and extended neuron somatic concepts that can be vastly used in creating such methods. Neural interrelations are described using graphs, construction of which is done in parallel with neural network learning. Prototype technique based on growing neural gas is also presented to give more detailed view.

Web textual documents scoring based on discrete transforms with fuzzy weighting

Recently, Fourier and cosine discrete transformations have been proposed for textual document ranking. The advantage of the methods is that rather than using only the count of a frequency term within document, the spatial information about presence of the term is considered. Here, further improvement of this novel approach is proposed. It is based on fuzzy evaluation of the position of words within the document.

Opinion Mining through Structural Data Analysis Using Neuronal Group Learning

Opinion Mining (OM) and Sentiment Analysis problems lay in the conjunction of such fields as Information Retrieval and Computational Linguistics. As the problems are semantic oriented, the solution must be looked for not in data as such, but in its meaning, considering complex (both internal and external,) domain specific context relations. This paper presents Opinion Mining as a specific definition of structural pattern recognition problem. Neuronal Group Learning, earlier presented as general structural data analysis tool, is specialised to infer annotations from natural language text.

Supervised Textual Document Classification Using Neuronal Group Learning

Together with fast development of different areas of pattern analysis, an increasing demand on new models and techniques is observed. Especially new information retrieval tasks, oriented on data meaning rather than layout, prove to be demanding for most known techniques. neuronal group learning concept presented in this article, together with prototype implementation gives flexibility of utilization of any kind of expert knowledge about the problem to ease classifier inference process. It can also be used to acquire structural knowledge about an object, which can later be used for solving a segmentation problem-often addressed in semantics-oriented text and image processing.