Container ISO code recognition system using multiple view based on Google LSTM tesseract

Abstract

Optical Character Recognition (OCR) system is vastly used to identify license plates, street signs, and other applications. However, it faces difficulties to recognize ISO code from natural images of shipping containers due to rough weather condition, varying color, illumination, etc. In this paper, these challenges were overcome by integrating deep learning-based OCR recognition from multiple view which increases both accuracy and reliability. Images are taken from three different views and based on the proposed algorithm it analyses all the images to detect ISO code format using sequence matching algorithm. Next, confidence level is calculated for each recognized code using Google LSTM neural-net based Tesseract engine model and identifies the one which has highest confidence level for the proposed OCR system to be robust in delivering high level accuracy in actual application.