Statistical pattern recognition for automatic writer identification and verification

The thesis addresses the problem of automatic person identification using scanned images of handwriting.Identifying the author of a handwritten sample using automatic image-based methods is an interesting pattern recognition problem with direct applicability in the forensic and historic document analysis fields.We describes a number of new and very effective techniques that we have developed in recent years for automatic writer identification and verification. Writer individuality is encoded using probability distribution functions (PDFs) extracted handwritten text blocks. Our methods operate at two levels of analysis: the texture level and the character-shape (allograph) level. At the texture level, we use contour-based joint directional PDFs that encode orientation and curvature information to give an intimate characterization of the individual handwriting style. In our analysis at the allograph level, the writer is considered to be characterized by a stochastic pattern generator of ink-trace fragments, or graphemes. The PDF of these simple shapes in a given handwriting sample is characteristic for the writer and is computed using a common codebook of graphemes obtained by clustering. There are two distinguishing characteristics of our approach: human intervention is minimized in the writer identification process and we encode individual handwriting style using features designed to be independent of the textual content of the handwritten sample. In our methods the computer is completely unaware of what has been written in the sample. The handwriting is merely seen as a texture characterized by some directional probability distributions or as a simple stochastic shape-emission process characterized by a grapheme occurrence probability. Our methods were statistically evaluated using datasets with handwriting samples collected from up to 900 subjects. The development of our writer identification techniques takes place at a time when many biometric methods (e.g. iris, fingerprint, face) undergo a transition from research to real full-scale deployment. Our methods also have practical feasibility and hold the promise of direct applicability. The main chapters of the thesis are based on a number of scientific papers that were published during the duration of the PhD project. Here is an outline with short abstracts for the main chapters and references to the corresponding papers.