Deep Learning-based Approach to OCR
SSCN Number : SSCN-0029
Project Number : FASKL_VKL2011127
Industry : Others
Solution : Machine Vision System/AIDC
Brands : Cognex
Problem Statement
The graphite electrode is labeled with a unique code for monitoring and tracking purpose. This unique code contains alphanumeric character that is stamped onto a graphite electrode. Poorly stamped characters, confusing background and unwanted noise make it difficult for a machine vision system with traditional rules-based approach to recognize the characters.
Solutions
With In-Sight ViDi, a powerful deep learning software platform, the In-Sight D900 vision system is able to read characters under challenging situations such as noisy backgrounds and poorly stamped characters. In the training phase, images under different conditions were collected and labeled for training and validation purposes. During training & validation phase, re-labelling of the incorrect characters or missed characters is done until the software’s model is able to correctly identifies all characters. When deployed, the In-Sight ViDi Read tool operates on input images and produces results based on the validated training data.
Benefits
Good reading rate for low quality ID stamped characters. With deep learning-based OCR, reading rate improves by at least 20% as compared to traditional rules-based algorithm.
