By the Power of Grayscale (zserge.com)

Grayskull is a guided tour and tiny C library that strips computer vision to the bare essentials: 8‑bit grayscale images stored as a flat uint8_t array and a single header of routines. The author shows how much useful CV functionality you can build without OpenCV or neural networks—using plain get/set helpers and simple loops—to target resource‑constrained devices, teach fundamentals, or produce highly predictable, low‑overhead image pipelines. The writeup walks through practical building blocks: image I/O as a WxH flat buffer, pixel ops (invert, mirror, crop), nearest‑neighbour and bilinear resizing (with float interpolation), convolutional filters (box, Gaussian, sharpen, emboss) and Sobel edge detection (computing gradient magnitudes). It covers segmentation via fixed, histogram/Otsu, and adaptive local thresholding, plus morphological operators (erosion, dilation, opening/closing) implemented as min/max over a radius. Small implementation details—kernels as signed 8‑bit arrays, normalisation of filter sums, computing Otsu’s inter‑class variance, and local averaging for adaptive thresholds—underscore that many CV tasks are simple numerical algorithms. For ML/vision practitioners this is a useful pedagogical reference and a practical toolkit for deterministic, low‑memory preprocessing on embedded systems where full libraries or deep models are overkill.