This set of lectures introduces several methods for producing compact descriptions of shapes, with the goal of producing descriptions that do not change (are invariant) to a variety of image and scene transformations. These descriptions can be used for recognizing objects or for matching locations or components to previously computed locations or descriptions.
This lecture introduces the concept that an image region or shape does not need to be described by listing all of the pixels that make up the shape. Instead, one can describe the shape using some numerical summaries. The lecture introduces some simple but robust region descriptions.
Moments are a family of shape descriptions that are compact, in that one can describe shapes consisting of thousands of pixels by simply a few numbers. The lecture also introduces methods so that the moments can be made invariant to the position that the object is in the image, or its orientation, or its scale.
This lecture looks at two techniques for describing a shape based on the properties of the boundary of an object.