Letters detection in the photograph

A system dedicated to visually impaired persons for revealing and reading aloud text may use standard optical character recognition (OCR) component. However the OCR component cannot analyze directly photographs taken out of doors in towns because of their complexity. Before launching the OCR component the photos require preprocessing that may relay on image segmentation and selection of letter-like segments. This approach is efficient especially in case of localizing short inscriptions consisting of one or two signs (e.g. tramway numbers). Its bottleneck is evaluation of image segment similarity to alphanumeric signs (i.e. letters and digits). This paper describes different methods and results of their comparison.