OpenCV

OpenCV is the fundamental library for Image Processing. v5.0 (2025) modernizes deep learning support and licensing.

When to Use

• Image Manipulation: Resizing, cropping, color space conversion (BGR -> RGB).
• Classic CV: Edge detection (Canny), Feature matching (SIFT/ORB).
• Video I/O: Reading/Writing webcams or video files.

Core Concepts

BGR

OpenCV reads images as Blue-Green-Red (not RGB) by default. History quirks.

cv::Mat

The core matrix structure (in C++). In Python, it’s just a NumPy array.

DNN Module

Running darknet/onnx models directly in OpenCV (lightweight inference).

Best Practices (2025)

Do:

• Use it for Preprocessing: cv2.resize() is highly optimized.
• Use headless: pip install opencv-python-headless for server deployments (smaller, no GUI deps).

Don’t:

• Don’t implement Deep Learning training: Use PyTorch. Use OpenCV only for inference/preprocessing.

References

• OpenCV Documentation