Document categorization and signature region analysis

This thesis contains studies related to automated analysis of document images. Two sub-problems in document analysis are considered for his purpose; automated categorization of documents and handwritten signature detection on documents. Classifying document images is an essential tool for many applications. This work presents a framework for categorizing documents which are frequently used in bank applications. The framework is based on the extracted text information and document image features. A feature extraction and selection technique is applied customized for Turkish texts. Categorization based on document image features is an alternative giving results in a faster way, because it works without the optical character recognition process, which is a computational intensive task. Automated localization of a handwritten signature in a scanned document is a promising facility for many banking and insurance related business activities. This work also describes here a discriminative framework to extract signature from a insurance service application document of any type. The framework is based on the classification of segmented image regions using a set of representative features. The segmentation is done using a two-phase connected component labeling approach. The combined effects of several feature representation schemes in distinguishing signature and non-signature segments is evaluated over a Support Vector Machine classifier. The experiments on a real insurance data set have shown that the developed framework can achieve a reasonably good accuracy to be used in real life applications.