Adaptive word style classification using a Gaussian Mixture Model

In this paper, we present a new approach to detect bold and italic words in scanned documents. Under the assumption that OCR results are available, features used for classification are selected automatically using feature selection. For each scanned page, a Gaussian Mixture Model is constructed for characters with the same character code, and word styles are determined using a weighted majority vote. We applied this method to a variety of documents and compared the results with current commercial OCR software that provides style information. The experimental results show that our method performs better.