Image Processing by Hand: Pixels, Convolution Kernels, Edges & Pooling
Treat images as numeric grids: greyscale from RGB, brightness, contrast and FREQUENCY histograms, thresholds and masks, 3×3 convolution with SUMPRODUCT, sharpening and Sobel edges, padding and stride, max and average pooling, and a tiny CNN from conv to probability, 8 graded labs with colour-scale heatmaps.
The first lesson is free — no account needed.
About this course
An image processing course that treats the pictures of one fictional photo app, PixelForge, as grids of numbers in a spreadsheet, for computer vision and machine learning learners, computer science students, teachers and curious developers. You convert colour pixels to greyscale with luma weights, brighten and stretch a dim image with clipping and draw its histogram with FREQUENCY, separate a coin from the background with a threshold and a binary mask, slide a 3×3 Gaussian kernel over an image with SUMPRODUCT, sharpen and detect edges with the Sobel kernels and the gradient magnitude, calculate output sizes for padding and stride and run a stride-2 convolution with OFFSET, shrink feature maps with max and average pooling, and push a handwritten digit through a tiny CNN from convolution to a sigmoid probability. Colour-scale heatmaps show every step. The first lab is free.
Lessons
- Pixels as Numbers: RGB to Greyscale with Weighted LuminanceFree
- Brightness, Contrast & the Histogram (FREQUENCY, with a Chart) Pro
- Thresholding & Binary Masks: Separating an Object from the Background Pro
- 3×3 Convolution with SUMPRODUCT: a Gaussian Blur Pro
- Sharpening & Edge Detection: Sobel Gx, Gy and Gradient Magnitude Pro
- Padding & Stride: the Output-Size Formula and a Stride-2 Convolution Pro
- Max & Average Pooling: Shrinking Feature Maps Pro
- A Tiny CNN by Hand: Conv → ReLU → Pool → Score Pro