Computer Vision Fundamentals: Image Formation, Filtering and Machine Learning by Hany Farid

Sobre este curso

Computer vision sits at the center of modern AI, powering everything from smartphone cameras and medical imaging to robotics and autonomous driving. This free online course builds a strong foundation by connecting how images are captured, how they can be transformed, and how learning algorithms can turn pixels into decisions. Instead of treating vision as a black box, you will develop an intuitive and practical understanding of why images look the way they do and how that influences the performance of machine learning systems.

You will start by exploring image formation from first principles: the pinhole camera model, perspective projection in 2D and 3D, and the coordinate transforms that map a world point onto a sensor. From there, you will bridge theory to real devices by looking at lenses, focal length, aperture, depth of focus, and exposure. You will also examine what a digital image really is at the pixel level, including color filter arrays, demosaicing, noise, chromatic aberration, lens distortion, displays, and what compression techniques like JPEG change in an image and why.

With that groundwork in place, you will learn how filtering works as a core tool for vision. Convolution is developed as a unifying idea across 1D and 2D signals, building up to concepts such as separable filters, edge and line detection, and descriptors that capture local structure. You will also gain a clearer understanding of the relationship between the spatial and frequency domains through Fourier representations and sampling, including why aliasing occurs and how it is mitigated in practice using smoothing and multi-scale representations like Gaussian and Laplacian pyramids.

The course then expands into image analysis and understanding: motion estimation and feature tracking, depth from stereo, epipolar constraints, and the geometric role of homographies in correcting perspective for planar surfaces. Finally, you will connect vision to supervised and unsupervised machine learning, studying regression and classification methods from least squares to logistic regression, LDA, SVMs, and neural networks, and seeing how optimization, generalization, and representation choices affect results. You will also encounter clustering, EM, PCA, and visualization techniques such as t-SNE, equipping you to reason about data structure as well as models.

By the end, you will be able to move confidently between physics, geometry, signal processing, and machine learning, making it easier to read research papers, debug vision pipelines, and design more reliable AI systems.