ErieTCG OCR
LANCER
Lancer:
<= 2 resources OR <=3 weeks effort OR <=$3000 USD
Growth:2 - 4 resources AND upto 4 months OR
upto $30000 USD
Enterprise:Everything else
(Development)
Project Description
ErieTCG OCR is an AI-powered engine that extracts structured data from trading card images, specifically targeting Basic-EX and Mega-EX cards. The application uses advanced image processing techniques and OCR capabilities to convert visual card data into searchable, structured formats. With modular architecture, custom object detection, and classification, the system streamlines card data management for collectors and platforms alike.
Scope Of Work
The client partnered with Oodles to create an automated image-to-text processing pipeline for trading cards. The objective was to build a microservice-based backend with a fully integrated OCR engine that identifies card attributes such as name, type, and number. The engine also needed to support image preprocessing, classification, custom Named Entity Recognition (NER), and deployment on a scalable cloud infrastructure.
Our Solution
To meet the project goals, Oodles delivered a scalable, end-to-end OCR pipeline with the following key features:
Modular Architecture and Secure APIs: Developed a microservice-based Python backend with a dedicated database and bearer token authentication for secure communication.
Image Processing and Data Preparation: Collected images of Basic-EX and Mega-EX cards, converted them to standard formats, applied noise-reduction techniques, and prepared annotated datasets for training.
Object Detection and OCR: Trained a YOLO model to identify key text regions and used PyTesseract and TrOCR to extract and structure the text data.
Classification and Text Refinement: Built a screen classification model and applied knowledge graph-based text preprocessing for improved accuracy.
Deployment and Handover: Deployed the solution using AWS ECS Fargate, integrated REST APIs, conducted thorough testing, and delivered full documentation to the client.