This page summarises a set of lectures that introduce computer vision to viewers who have a basic scientific, and mathematical and computational background. They present the basic concepts behind how images are created from light, how images are captured by cameras, and what sorts of problems can arise because of the imaging process. Then, the lectures look at how information can be extracted from the images, from both static and active sensors.
This course segment gives a brief introduction to the content of the five main blocks of the lecture series: image and capture, how to isolate structure in an image, how to describe the isolated structure, how to recognise objects with those descriptions and what extra information is available if you have a sequence of images, e.g. from video.
This set of lectures introduces what an image consists of, from a computer's point of view, how the computer can capture the image, what can go wrong, some of the different types of information in an image, what makes image analysis hard, and the mathematics and geometry of image projection. Then, exploiting the geometry and mathematics, we introduce the concept of homography and show how one can use a homography to map image data from one image to another.
Analysing the content of an image may require isolating the important regions or structures in the image. These lectures introduce some of the approaches to segmentation, namely thresholding, mean shift segmentation and background comparison with moving objects in video.
Using the raw pixel data is normally too computationally demanding, so one typically uses descriptions of the segmented structures instead of the structures themselves. These lectures present several methods for describing structures, most notably image moments and signatures.
A common task in the area of image analysis is recognition, whether of faces, places or objects. These lectures introduce a Bayesian probabilistic approach and a geometric template matching approach to object recognition.
One can use a sequence of images as a set, rather than as single images. This allows computation of where important changes are taking place in a scene, or what has changed, or what the geometric structure of the scene is. These lectures introduce these ideas.
This brief lecture summarises the five main topics from the lecture series and gives some directions for further study about computer vision.