Computer vision

Classical image processing and geometry first, then convolutional networks and object detection, so the learned methods are seen against what they replaced.

This part builds on Signal processing and Neural networks.

2 of 15 topics ready, listed in reading order

Digital images and colourRaster and vector images, colour spaces and alpha compositing.Planned

Point operations and histogramsContrast, gamma and histogram equalization.Planned

Convolution and filteringLinear and non-linear filters, smoothing, sharpening and gradients.Planned

Edge detectionThe Canny detector step by step.Planned

MorphologyErosion, dilation, opening and closing on binary and grey images.Planned

Geometric transformsResampling, affine transforms and homographies.Planned

SegmentationOtsu thresholding, clustering, superpixels and a promptable foundation model for comparison.Planned

Contours and shapeContour tracing and shape measurements.Planned

Features and matchingHarris corners, scale space and SIFT, template matching, RANSAC and panorama stitching.Planned

Face detectionIntegral images, Haar features, Viola-Jones and HOG with an SVM.Planned

Motion and trackingBackground subtraction, frame differencing and simple tracking.Planned

Camera models and calibrationThe pinhole model, intrinsics, distortion and calibration.Planned

Convolutional networksConvolution arithmetic, parameter and FLOP budgets, pooling and lightweight blocks.Ready

Training image classifiersClean evaluation, augmentation, class imbalance and dataset shift.Planned

Object detectionIoU, non-maximum suppression, anchors, YOLO-style decoding, the R-CNN family and mAP.Ready