Abstract: The talk presents an overview of the approaches necessary to build
the best commercial OCR technology. Commercial OCR technology vendors
face significant challenges that academy researchers don't have to take care
of. The more popular a technology is, the more important it is to keep
recognition results consistent among different versions of the technology SDK
in all usage scenarios it is being employed, regardless of any improvements in
architecture and algorithms. OCR technology has much more quality
parameters than just recognition accuracy, and in different usage scenarios
different parameters are important. Thus the best technology has to be best
in all aspects. Moreover, during past decades industry has redefined the term
OCR which now means much more that just character recognition. This talk
attempts to discuss these challenges and their influence on the research and
development process. |
