Challenges in Input Preprocessing for Mobile OCR Applications: A Realistic Testing Scenario

Applications of OCR range from automatic text input from images and videos and text-to-speech to automatic translation of textual information in real world scenes. Such wide applicability of OCR has promoted its use on mobile devices, such as smartphones. However, there are intrinsic complications when designing an end-to-end OCR system for smartphones, as there is a number of steps and challenges, inherent to OCR itself, which may prove arduous and inconvenient for a conventional mobile device to handle. For starters, before actual OCR, input needs to be preprocessed, in order to enhance the existing textual information and minimize noise and outliers. In this study, we attempt to document the challenges faced by a smartphone-enabled OCR application, experiment on the preprocessing steps using a realistic testing scenario and pose some open issues for future investigation.