Computers today are limited in their ability to interact with the world and with their human users because they lack the ability to "see". The study of computer vision requires that we understand something about the physics of the world, how light is reflected off surfaces, how objects move, and how all of this information gets projected onto an image by the optics of a camera. It also requires that we devise algorithms to recover, or reconstruct, some of these physical properties from one or more images. This "inverse" problem is a great puzzle. Information is lost when the three dimensional world is projected onto a two dimensional image; how can we recover this information from a picture of it? The course introduces the algorithms behind this and develops methods for solving various inverse problems. But vision is about more than simply reconstructing the 3D world from 2D images; it is about extracting semantics. The course will explore machine learning techniques and probabilistic inference methods that begin to address this problem. In this course you will
- be exposed to many areas of current computer vision research
- implement a number of programming assignments to get hands-on experience working with images and image sequences
- find out that all that linear algebra and calculus you learned is actually useful for something real
Even if you do not go on to study computer vision, the basic tools and techniques we use here will be useful in many other areas. For all others that would like to study Computer Vision in more details, this course offers you the possibility to find out which sub topic would be most interesting to you.
How can computers understand the visual world of humans?
This course treats vision as a process of inference from noisy and uncertain data and emphasizes probabilistic and statistical approaches. Topics include: Structure of Images, Texture, Scenes, and Context, Feature Based Alignment, Stereo Correspondence, Structure from Motion, Image Stitching, Computational Photography, Image Based Rendering, and Baysian Modeling for Object Recognition. This course is designed for students interested in vision, artificial intelligence, or machine learning. The course offers a broad introduction to the field, the current problems and theories, the basic mathematics, and some interesting algorithms and the possibility to apply the contents learned direct in exercises.