Introduction to Computer Vision with OpenCV
summary
OpenCV techniques are used to replace slow deep models for document and ID-image preprocessing: a pipeline combining grabCut, erosion/dilation, minimum-area rectangles and feature/template matching deskews and reorients driver’s licenses, reducing runtime from ~7s to ~1s and supporting 9 new plus 8 legacy variants. For email document classification, dilating text and extracting contours yields an average text angle to correct rotation and a small-template check fixes flipped orientation, markedly improving OCR quality and throughput; optional deep models can boost recall but require a GPU, and many components can be compiled to WebAssembly for in-browser use.
AI-written summary · about
chapters (18)
- 0:00Computer Vision with OpenCV intro
- 1:28Real-time face detection
- 3:16Image basics and pixels
- 6:02Thresholding and erosion/dilation
- 7:49Edge detection
- 8:31GrabCut object extraction
- 9:14Feature matching for object detection
- 10:38Template matching comparison
- 12:02Text detection with OpenCV
- 12:44Driver's license use cases
- 15:55Image rectification demo
- 17:43Old license rectification
- 20:07Automatic email sorting
- 21:54Text rectification demo
- 22:56Text detection with Deep Learning
- 24:02OpenCV in JavaScript and WebAssembly
- 25:47Final demo and conclusion
- 26:47Q&A
related talks
01h 13mChallenges and Trade-offs for In-Browser Computer Vision Applications
26mChallenge Accepted - How to Escape the Quicksand While Engineering a Computer V...
42mUsing Deep Learning to Understand Documents
20mComputer Vision at the Edge with OpenVINO
37mComputer Vision Using Go And OpenCV 3
41mComputer Vision with GO and OpenCV
01h 12mBuilding a live face recognition system
42mComputer Vision Using Go And OpenCV 3