Natalia Vassilieva

Image Retrieval: Color and Texture Combining Based on Query-Image

It is a common way to process different image features independently in order to measure similarity between images. Color and texture are the common ones to use for searching in natural images. In [10] a technique to combine color and texture features based on a particular query-image in order to improve retrieval efficiency was proposed. Weighted linear combination of color and texture metrics was considered as a mixed-metrics. In this paper the mixed-metrics with different weights are compared to pure color and texture metrics and widely used CombMNZ data fusion algorithm. Experiments show that proposed metrics outperform CombMNZ method in some cases, and have close results in others.

Robust versions of the Tukey boxplot with their application to detection of outliers

The need for fast on-line algorithms to analyze high data-rate measurements is a vital element in production settings. Given the ever-increasing number of data sources coupled with increasing complexity of applications, and workload patterns, anomaly detection methods should be light-weight and must operate in real-time. In many modern applications, data arrive in a streaming fashion. Therefore, the underlying assumption of classical methods that the data is a sample from a stable distribution is not valid, and Gaussian and non-parametric based methods such as the control chart and boxplot are inadequate. Streaming data is an ever-changing superposition of distributions. Detection of such changes in real-time is one of the fundamental challenges. We propose low-complexity robust modifications to the conventional Tukey boxplot based on fast highly efficient robust estimates of scale. Results using synthetic as well as real-world data show that our methods outperform the Tukey boxplot and methods based on Gaussian limits.

A New Measure of Outlier Detection Performance

Traditionally, the performance of statistical tests for outlier detection is evaluated by their power and false alarm rate. It requires ensuring the upper bound for false alarm rate while measuring the detection power, which proves to be a difficult task. In this paper we introduce a new measure of outlier detection performance H m as the harmonic mean of the power and unit minus false alarm rate. The H m maximizes the detection power by minimizing the false alarm rate and enables an easier way for evaluation and parameters tuning of an outlier detection algorithm.

Texture-Based Text Detection in Digital Images with Wavelet Features and Support Vector Machines

In this paper we propose to combine region-based and texture-based approaches for text detection in digital images. Our solution is based on a cascade filtering of image regions. First, we apply heuristic filtering to disregard certain non-textual areas. Second, we perform a more precise and expensive texture-based filtering using support vector machines and wavelet-based texture features. We have evaluated our approach with the ICDAR 2003 text locating competition benchmark collection and tools. The experimental results showed competitive performance of our solution by means of recall and precision compared to other text detection approaches participated in ICDAR 2003 and lower computational cost at the same time.

High-recall extraction of acronym-definition pairs with relevance feedback

This paper addresses the problem of extracting acronyms and their definitions from large documents in a setting, when high recall is required and user feedback is available. We propose a three step approach to deal with the problem. First, acronym candidates are extracted using a weak regular expression. This step results in a list of acronyms with high recall but low precision rates. Second, definitions are constructed for every acronym candidate from its surrounding text. And last, a classifier is used to select genuine acronym-definition pairs. At the last step we use relevance feedback mechanism to tune the classifier model for every particular document. This allows achieving reasonable precision without losing recall. As opposed to existing approaches, either created to be generic and domain independent or tuned to one particular domain, our method is adaptive to an input document. We evaluate the proposed approach using three datasets from different domains. The experiments prove the validity of the presented ideas.