Levensthein distance as a post-process to improve the performance of OCR in written road signs
Satria Priambada, Dwi Hendratmo Widyantoro · 2017
Recognizing characters in a picture is one of the image processing problem that has resulted in a very high accuracy of character detection. One technique is using OCR with library Tesseract. However, this method has been known to produce a poor result to recognize character inside a video. In this paper we attempt to improve the performance of OCR with Tesseract in by correcting OCR results during post processing. The Post process uses some rules to omit non possible characters. A Levensthein Distance is then applied to find a road name and location in a database with the minimum distance to the OCR reading. The experiment is conducted with Indonesian written road signs. The prototype can run in real-time for street signs information extraction with average processing time of each frame around 300-600 ms. Our evaluation reveals that applying Levensthein Distance during post processing stage can improve precision, recall, and F-measure from 17% to 60%.