Challenge Accepted - How to Escape the Quicksand While Engineering a Computer V...
summary
A working computer vision pipeline maps a phone or webcam frame to a PDF page via feature-matching to PDF pages, detects highlighted regions using a mix of low‑level image processing and frame differencing with optical flow to ignore hands, and extracts text by falling back from parsed PDF to OCR then postprocessing, summarized as Tesseract + lightweight PyTorch reclassifier to recover superscripts and subscripts. Evaluation uses fast feedback loops on curator-supplied images, and initial experiments show SIFT too slow so faster feature detectors are profiled for near-real-time use.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:10What is a biomedical curator
- 6:48Pipeline overview
- 8:32Challenges
- 9:43Illumination Conditions
- 12:05Text Extraction
- 16:17Benefits of video data
- 17:28How does it work
- 18:41Do curators add labels
- 20:25Differences between images
- 21:55Reference mapping
- 22:59Languages
- 23:50Conclusion
related talks
30mIntroduction to Computer Vision with OpenCV
33mComputer Vision, Not Just for Breakfast Anymore
41mPractical Aspects of Applying Modern Computer Vision Technologies
20mScalable OCR pipelines using Python,Tensorflow & Tesseract
22mOCR, Recognition and Information Extraction from Images
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
41mKEYNOTE: Image processing, AI, and autonomous systems
38mDeep Learning for detection on a phone