Sebastian Klenk

Robust 1D Barcode Recognition on Mobile Devices

In the following we will describe a novel method for decoding linear barcodes from blurry camera images. Our goal was to develop a algorithm that can be used on mobile devices to recognize product numbers from EAN or UPC barcodes.

The Normalized Compression Distance as a Distance Measure in Entity Identification

The identification of identical entities accross heterogeneous data sources still involves a large amount of manual processing. This is mainly due to the fact that different sources use different data representations in varying semantic contexts. Up to now entity identification requires either the --- often manual --- unification of different representations, or alternatively the effort of programming tools with specialized interfaces for each representation type. However, for large and sparse databases, which are common e.g. for medical data, the manual approach becomes infeasible. We have developed a widely applicable compression based approach that does not rely on structural or semantical unity. The results we have obtained are promising both in recognition precision and performance.

Interactive survival analysis with the OCDM system: From development to application

Medical data mining is currently actively pursued in computer science and statistical research but not in medical practice. The reasons therefore lie in the difficulties of handling and statistically analyzing medical data. We have developed a system that allows practitioners in the field to interactively analyze their data without assistance of statisticians or data mining experts. In the course of this paper we will introduce data mining of medical data and show how this can be achieved for survival data. We will demonstrate how to solve common problems of interactive survival analysis by presenting the Online Clinical Data Mining (OCDM) system. Thereby the main focus is on similarity based queries, a new method to select similar cases based on their covariables and the influence of these on their survival.

Interactive feature visualization for image retrieval

Most systems for content based image retrieval (CBIR) employ low level image features as a similarity measure. The problem of CBIR systems is that they are a ldquoblack boxrdquo to the user: Queries are specified by sample images, but the features which the CBIR system actually uses are unknown to the user. Hence, unexpected results are difficult to interpret. The problem becomes worse for inexperienced users, who expect the system to understand their query on a symbolic level, while in reality the CBIR system just extracts close-to-signal features. Here we propose to make CBIR systems more ldquotransparentrdquo by visualization of the employed features. Since non-experts should be able to operate the CBIR system, we argue that features should be visualized as prototypical, artificial images, rather than feature-specific visualizations (such as bar-diagrams for a histogram). We present the visualization of two widely used feature classes, color histograms and texture features, and evaluate in a user study how well the visualizations can be interpreted.

Redundant dictionary spaces as a general concept for the analysis of non-vectorial data

Many types of data we are facing today are non-vectorial. But most of the analysis techniques are based on vector spaces and heavily depend on the underlying vector space properties. In order to apply such vector space techniques to non-vectorial data, so far only highly specialized methods have been suggested. We present a uniform and general approach to construct vector spaces from non-vectorial data. For this we develop a procedure to map each data element in a special kind of coordinate space which we call redundant dictionary space (RDS). The mapped vector space elements can be added, scaled and analyzed like vectors and thus allows any vector space analysis techniques to be used with any kind of data. The only requirement is the existence of a suitable inner product kernel.

A personalized medical information system

Shared decision making is not just a question of collecting enough information, but mostly of gathering the right information and evaluating it correctly. The ever growing amount of available data and the constantly increasing specialization in medicine makes it almost impossible for a patient to get personalized medical information even though this is crucial for a self determined decision. We therefore propose a system that combines epidemiological data, personal medical data, personal data and publicly available data to form one central source of information. We argue that, with currently available methods and data, a patient adapted information system is attainable.

Relevance based visualization of large cancer patient populations

Cancer patient data is usually visualized in an aggregated fashion -- e.g., Kaplan-Meier diagrams show the average survival estimate, but leave the viewer uninformed about any special cases. Work with large patient data corpora, e.g. medical web data, often requires both information about the whole corpus as well as detailed information about a single case. The latter is particularly important for the analysis of outliers. We present a method to visualize high dimensional cancer patient data in the form of a two dimensional scatter plot such that both a large scale overview is given and at the same time detailed information about every single patient is displayed. As a projection of high dimensional data onto a space of much lower dimension is bound to reduce information, our method allows to select the most important parameter (survival time) to be preserved in the projection. We present the algorithm and use it to visualize breast cancer patient data. We show the visualizations together with the resulting relevance vectors for an in-depth study.