Scalable OCR pipelines using Python,Tensorflow & Tesseract
summary
Production OCR pipeline that combines OpenCV pre-processing (binarization, Hough-line rotation, dilation/erosion for text segmentation) with Tesseract OCR at scale inside Docker containers, plus an Elasticsearch-backed fuzzy-correction dictionary. TensorFlow trains a CNN classifier to filter images from text (claimed 100% on evaluation) and uses Grad-CAM to inspect failures. Results: 80% of pages reach 100% recognition after preprocess, Tesseract costs ~12s per 300 DPI image, and architecture targets ~140,000 images/day with sharded Elasticsearch and containerized workers.
AI-written summary · about
chapters (5)
- 0:00Start
- 3:12What Is Binarization
- 13:49How Do We Get the Recognition Rate
- 16:06Scalability
- 16:17Scale the Ocr Process
related talks
22mOCR, Recognition and Information Extraction from Images
41mOCR in the Wild World of Social Media
28mScalable High Precision OCRs for Game Event Detections
15mMachine Learning with TensorFLow Object Detection running on Docker
26mChallenge Accepted - How to Escape the Quicksand While Engineering a Computer V...
19mDocument Reader using Deep Learning with a Small Dataset
41mApplying serverless architecture pattern to distributed data processing
15mFrom Image to Graph: Leveraging Tesseract OCR Engine for Document Chunking and GraphRAG