Long before neural networks, vision systems were built from hand-designed operations on pixels. These methods are still everywhere: they clean up images before a model sees them, they power fast tools like panorama stitching, and they explain what deep networks later learned to do on their own.
Filtering slides a small grid of numbers, a kernel, across an image to blur it or sharpen it, an operation called convolution. Edge detection uses the same idea to find where brightness changes sharply, with Sobel and Canny as the classic methods. Local features such as SIFT and ORB pick out distinctive points that can be matched between two images.
Finally, camera geometry explains how a 3D scene lands on a 2D image, through the pinhole model and the intrinsic and extrinsic camera parameters. Watch the lessons in order, then continue on to Deep Learning Foundations.
How kernel convolution produces mean and Gaussian blurs, and how more complex filters are built by chaining simple operations together.
How the Canny detector builds on Sobel gradients, thins edges to a single pixel, and uses two thresholds to keep only the strongest outlines.
A short explainer on SIFT, the Scale-Invariant Feature Transform, and how it finds features that can be matched between images.
The two sets of camera parameters: intrinsics inside the camera, and extrinsics for its position and orientation in the world.