Comparison between Pen-scanner and Digital Camera Acquisition for Engraved Character Recognition

Engraved characters are usually a limitation for character recognition, as their extraction is difficult due to the small contrast with the background. Nevertheless, they are present in several industrial applications and their handling may be convenient for special tasks such as traceability or quality control. We propose here an entire system from acquisition to character recognition by considering two different and mobile acquisition devices, a pen-scanner close to traditional acquisition and a digital camera in order to appreciate performance and differences. In some industrial applications such as the ones including engraved characters, it may be difficult and awkward to properly build a large training database for recognition. Hence, we also propose a novel method to increase automatically the size of the training dataset without loosing representativeness by using image analogies. Results are finally presented in terms of Precision and Recall for character extraction quality and recognition rates for character recognition with comparison with a commercial OCR.