Punjabi Text Recognition System for Portable Devices: A Comparative Performance Analysis of Cloud Vision API with Tesseract

Abstract

The increasing availability of high performance, low priced, portable digital imaging devices has created an opportunity for on demand analysis of documents. In this paper, Punjabi Text Recognition System is developed for portable devices using two different approaches that is Google’s Cloud Vision APIs and LSTM based Tesseract OCR Engine. The performance of developed mobile based systems is compared in term of runtime and recognition accuracy. Both Vision API and LSTM based OCR engine provides good results for Roman Based Scripts. Particularly for Gurmukhi text document images, Cloud Vision API recognizes Punjabi with good accuracy as compared to Tesseract. We presented a detailed comparison and computed the character and word level accuracy of both the systems for same set of images.