An image is a function.
Everything else is what you do to it.
A from-scratch reference on digital image processing — built to go deep on the fundamentals and even deeper on the specific problem of matching LiDAR intensity images against reference DEMs and optical imagery. Live demos throughout: drag sliders, hover pixels, watch transforms happen.
§ How this is organized
The material runs in the conventional order this field is usually taught in: how images come to exist as sampled, quantized arrays; how to reshape their intensities and neighborhoods; how to see them as sums of frequencies instead of grids of pixels; how to undo damage, reshape geometry, and eventually describe what's in them well enough to match one against another. The last five chapters exist because of one specific problem — matching a LiDAR intensity return against a reference DEM or orthoimage, invariant to rotation and translation, fast enough to run on an FPGA — and go deeper than a general course would.
The image as signal
Sampling, quantization, intensity transforms, convolution, and the 2D Fourier transform.
Repair & reshape
Noise models and restoration filters, morphology, geometric transforms and interpolation.
Finding structure
Edges, thresholding, region growing, and describing boundaries and regions numerically.
Project track: registration
Harris / ORB / SIFT, phase correlation, multimodal LiDAR–DEM–optical matching, CNN features, FPGA quantization.
These notes were drafted with the help of Claude (Anthropic) as a study aid, then reviewed and organized by me. Explanations draw on general, first-principles knowledge of the field rather than being transcribed or paraphrased from any specific textbook, and the demos were built for this reference. Treat it as a personal study companion, not a replacement for course material or the primary literature.