Description: The problem of highly imbalanced datasets with only sparse data of the minority class in the context of twoclass
classification is investigated. A novel synthetic data oversampling technique is proposed which utilizes
estimations of the probability density distribution in the feature space. First, a Gaussian mixture model (GMM)
from the data of the well-sampled majority class is generated and with its help a new GMM is approximated by
Bayesian adaptation using the sparse minority class data. Random synthetic data is generated from the adapted
GMM and an additional assignment rule assigns this data to either the minority class or else discards it. The
obtained synthetic data is employed in combination with the available original data to train a support vector
machine classifier. The examined application in this paper is optical on-line process monitoring of laser brazing
with only rare sporadic occurring defects. Experiments with different amounts of minority class data samples

Description: The document preprocessing output may affect the efficiency of the document analysis and recognition system. In order to ameliorate the efficiency of those systems, an objective evaluation of preprocessing step is essential. In this paper we propose a new framework for automatic evaluation of binarization approaches. The evaluation of binarization approaches is based on the comparison between input images and generated ground-truth. The automatic method of ground-truth generation is an improvement of a semi-automatic method. The tests was applied on two different datasets, training dataset is composed from parts of documents, the test dataset is composed from entire historical documents.   

Description: Historical documents contain generally different kind of degradations. Due to this degradations the application of methods of noise removal during a preprocessing stage seems to be necessary. Since the noise which, exists in the original document can not be eliminated using a simple noise removal algorithm and it influences the preprocessing result e.g. the binarization, a function of noise detection seems to be necessary. We present in this paper a method for the selection of the input parameters of binarization methods according to the noise type detected in the image. The tests are achieved on benchmarking datasets used at DIBCO 2009 and H-DIBCO 2010. The results returned by the binarization methods using the noise features are promising.