Skip to Main Content
Mobile platforms usually have highly versatile resources. To be a successful interface solution for mobile devices, OCR should have high computational efficiency and adaptation to platform diversity, which are usually trivial in a desktop system, in addition to good recognition rate. In this paper, a practical case study is presented, in which an OCR software is reformed from desktop to embedded version for mobile platforms. To reduce computational burden, all calculations are integerized and math functions are implemented in piecewise linear approximation. An idea of data structure for plug-in template is devised to enhance portability. A complete practical re-coding work has been done. In it a desktop version neural network based OCR is converted into an embedded style. The test results show 60% speed up without sacrificing the correct recognition rate, running on a half sized executable code.