Today, advances in Computer Science and the proliferation of computers in modern society are an unquestionable fact. Nevertheless, the continuing importance of orthography and the hand-written document are also beyond doubt. The new technologies permit us to work with online information collecting, but there is still a large quantity of information in our society which requires using algorithms for samples off-line. Security in certain applications requires having biometric systems for their identification; in particular, banking checks, wills, postcards, invoices, medical prescriptions, etc, require the identity of the person who has written them to be verified. The only way to do this is with writer recognition techniques. Furthermore, many hand-written documents are vulnerable to possible forgeries, deformations or copies, and generally, to illicit misuse. Therefore, a high percentage of routine work is carried out by experts and professionals in this field, whose task is to certify and to judge the authenticity or falsehood of handwritten documents (for example: wills) in a judicial procedure. Therefore nowadays research on writer identification is an active field. At present, some software tools enable certain characteristics to be displayed and visualised by experts and professionals, but these experts need to devote a great deal of time to such investigations before they are able to draw up conclusions about a given body of writing. Therefore, these tools are not time-saving and nor do they provide a meticulous analysis of the writing. They have to work with graph paper and templates in order to obtain parameters (angles, dimensions of the line, directions, parallelisms, curvatures, alignments, etc.). Moreover, they have to use a magnifying glass and graph paper in order to measure angles and lines. This research aims to lighten this arduous task.
Encoding CNN Activations for Writer Recognition