Computer Vision: Algorithms and Applications, 2nd Edition

What this book is

Computer Vision: Algorithms and Applications is the textbook many university vision courses use, written by Richard Szeliski. The second edition was published by Springer in 2022. The author's site offers the PDF free for personal use; he asks that it is not reposted elsewhere. A printed copy is available from the usual booksellers.

What it covers

The chapters run from introduction and image formation, through image processing and model fitting and optimisation, to deep learning and recognition. They continue with feature detection and matching, image alignment and stitching, motion estimation, computational photography, structure from motion and SLAM, depth estimation, 3D reconstruction and image-based rendering, with appendices on linear algebra and Bayesian modelling. The second edition moved deep learning and recognition earlier, reflecting how modern systems are built, and its recognition chapter includes semantic segmentation, video understanding and vision and language.

Is it worth your time?

As a deeper reference, yes. It is not a coding tutorial and you do not need to read it cover to cover. Its value is breadth and precision: when you want to understand why a detector struggles with small objects, how cameras form images, how panoramas are stitched or how 3D reconstruction works, it explains the ideas properly and points to the key papers.

How to use it

Read the deep learning and recognition chapters alongside a practical course, then dip into feature matching, alignment or 3D chapters when a project needs them. The image formation and image processing chapters are worth reading early, because they explain much of what goes wrong with real photos.

Prerequisites

Comfortable linear algebra, calculus and probability, plus some programming. It is written for senior undergraduates and graduate students.

Free and paid

The PDF is free for personal use from the author's site. The printed book is paid.

Common questions

Is it outdated by deep learning? No. The second edition integrates deep learning throughout, and the classic geometry and image processing chapters remain the foundation for 3D vision, robotics and camera work.

Best for: learners who have trained a few vision models and want the full picture of the field.