Development of OCR-based Applications to Improve Inventory Management

Abstract

Background/Objectives: Inventory management is essential in today's business. Accurate inventory tracking allows company save their time and money for quick and accurate inventory scanning, we suggest OCR-based inventory management software.

Methods/Statistical analysis: The process go through 5 steps: pre-processing, text detection, text classification, text recognition, post-processing. We adopted character-level text detection method instead of word-level's. The annotation of character-level is extracted from word-level's through weakly supervised learning.

Findings:  In this way, we experimented with various previous character recognition methods with the same data set to compare performance. Furthermore, we have combined each method, and tested it with various data sets. The results showed generally higher performance than conventional methods. Unlike other image processing tasks that require RGB channels, we have found out gray scaling image in OCR process brings high accuracy of character recognition. You can check this on Table1. We have made the network spatially invariant without data augmentation by using STN(Spatial Transformer Network).

Improvements/Applications:  The use of OCR technology has been confined to limited fields such as document translation or license plate recognition due to it's performance. We have improved the accuracy of text recognition by combining the existing OCR methods. Based on our results, we try to increase the use of OCR technology to inventory management system.

Keywords:  OCR, Attention, Inventory Management, STN, Text Detection, Text Recognition