Computer vision is concerned with developing artificial systems which extract information from image or video data. It is a hybrid subject at the junctions between neuroscience, machine learning, geometry, mathematics and electrical engineering.

Automated vision is the opposite process to computer graphics: the aim is to construct a model of the 3d scene from images rather than vice‐versa. Unfortunately, the mapping from scenes to images is many to one: there may be several possible configurations of the world that produce exactly the same image. This is one of the reasons vision is challenging.

Vision is also difficult because of the sheer variation in appearance of objects. For example, when we compare images of two dogs (e.g. a poodle and a great dane), the actual pixel values have very little in common. However, humans have no problem recognizing both images as examples of dogs and distinguishing them from images of cats: we know vision is possible as the human visual system provides an existence proof. In fact more than one third of the brain is involved in vision, suggesting that considerable computational resources are required.

Vision tasks include: reconstruction in which we build a three dimensional model of the scene from one or more images (right), camera tracking where we identify the movement of the camera relative to the scene, object detection in which we determine if a certain type of object (e.g. a dog) is in the image and segmentation (above right) in which we divide the image up into meaningful regions. Applications of machine vision include robotics, face recognition, content based image and video retrieval, building 3d models from photos and many industrial applications.

BACK TO MSC CGVI OVERVIEW