An object location strategy using shape and grey-level models

Abstract

This work forms part of a project which is directed towards producing a computer vision system which may be used by a person who is not familiar with image processing techniques to define tasks and produce solutions to 2-D image interpretation problems. The proposed system models image data and allows a user to describe an image processing task in an hierarchical manner in what is collectively termed a world model 1 . The purpose of the hierarchy is to split the task and the data modelling into manageable units, to assist in reducing the combinatorial problems encountered in model-based image interpretation 4 . Execution of the task is synonymous with instantiation of the model from image data.

Boundary Detection Using Bayesian Nets

“…Many image analysis tasks can be carried out by making inferences from the evidence available in 1-D profiles taken from selected parts of the image, combined with constraints defined by the user, and possibly estimated from a training set [1,2]. A fundamental task is thus to label segments of the profile as being "explained" by the presence in the scene of objects drawn from predefined classes.…”