Adolfo Santoro

Writing Order Recovery from Off-Line Handwriting by Graph Traversal

We present a method to recover the dynamic writing order from static images of handwriting. The static handwriting is initially represented by its skeleton, which is then converted into a graph, whose arcs correspond to the skeleton branches, and nodes to either end point or branch point of the skeleton. Criteria derived by handwriting generation are then applied to transform the graph in such a way that all its nodes, but the first and the last, have an even degree, so that it can be traversed from the first to the last by using the Fleurys algorithm. The experimental results show that combining criteria derived from handwriting generation models with graph traversal leads to reconstruct the original sequence produced by a writer even in case of complex handwriting, i.e handwriting with retracing, crossings and pen-ups.

Modeling Handwriting Style: A Preliminary Investigation

We present a study for modeling handwriting styles that derives from handwriting generation studies, according to which handwriting is a temporal sequence of elementary movements. Hence, handwriting style results from the way those movements are actually performed and sequentially executed to reach fluency. We conjecture that handwriting styles depend on two main factors: the shape of the traces corresponding to the elementary movements and the way these traces are connected. To prove this conjecture, and the handwriting style model we have derived from it, we have designed an experiment in which handwriting samples are described by only two parameters and then clustered. The experimental results show that, despite its simplicity, the proposed method is able to capture the distinctive aspects of handwriting styles behind the handwriting samples, even when the writers deliberately attempts to modify it, and therefore corroborate our conjecture.

Modelling Visual Appearance of Handwriting

We present an experimental validation of a model of handwriting style that builds upon a neuro-computational model of motor learning and execution. We hypothesize that handwriting style emerges from the concatenation of highly automated writing movements, called invariants, that have been learned by the subject in correspondence to the most frequent sequence of characters the subject is familiar with. We also assume that the actual shape of the ink trace contains enough information to characterize the handwriting style. The experimental results on a data set containing genuine, disguised, and forged (both skilled and naive) documents show that the model is an effective tool for modeling intra-writer and inter-writers variability and provides quantitative estimation of the difference between handwriting styles that is in accordance with the difference in the visual appearance of the handwriting.

An Interactive Tool for Forensic Handwriting Examination

We introduce a tool for quantitative evaluation of handwriting features largely adopted during forensic examination of questioned documents. The tool is based on a model of handwriting generation and execution according to which handwriting is composed of elementary movements, called strokes, whose order and timing of execution has been learned and stored in the brain. Thus, what characterizes handwriting individuality, and therefore should be inferred from the samples available, is the way the sequence of strokes are executed. The tool does not aim at reaching a conclusion on the writers identity when comparing two documents, but provides the quantitative evaluation of a set of features that can be used by the expert to support his/her conclusion. Although the tool is meant to proceed automatically from the scanned image of the document to the quantitative evaluation of the features, it is equipped with an interface that allows the expert to follow the automatic procedure step-by-step and even to modify the output of any step and to modify it in case it is deemed as incorrect. The tool automatically produces a customizable report to illustrate the procedure, the features computation and to show the computed features values in both numerical and graphical form.

Writer Verification in Forensic Handwriting Examination: A Pilot Study

Writer verification in forensic handwriting examination is usually performed on just a few lines of text available in both genuine and questioned documents. In such a case, the only source of information being the feature values measured on the document at hand, the large majority of the methods proposed in the realm of pattern recognition cannot be applied, as the large set of samples they require for training is not available. In this study we investigate to which extent a statistical characterization of the variability exhibited by a set of features commonly adopted by forensic handwriting examiners may provide valuable clue to weight the evidence drawn from the available data. We argue that such a characterization should be given in terms of parameters that can be directly linked to the perceived variation between documents, so as to be easy to understand and to report in courts. The results of a set of experiments show that simple statistical parameters, such as mean and standard error of the mean, may provide an effective way to establish both the reliability of a feature for the case at hand, as well as its relevance for verification.

A Human in the Loop Approach to Historical Handwritten Documents Transcription

We propose a novel approach for helping content transcription of handwritten digital documents. The approach adopts a segmentation based keyword retrieval approach that follows query-by-string paradigm and exploits the user validation of the retrieved words to improve its performance during operation. Our approach starts with an initial training set, which contains only a few pages and a tentative list of words supposedly in the document, and iteratively interleaves a word retrieval step by the system with a validation step by the user. After each iteration, the system exploits the results of the validation to update its internal model, so as to use that evidence in further iterations of the search. Experimental results on the Bentham dataset show that the system may start with a few word images and their transcripts, exhibits an improvement of the performance during operation, and after a few iterations is able to correctly transcribe more than 68% of the word of the list.