One of the fundamental problems in the field of computer vision is the task of classifyingudobjects, which are present in an image or sequence of images, based on their appearance.udThis task is commonly referred to as the object recognition problem. A system designed toudperform this task must be able to learn visual cues such as shape, colour and texture fromudexamples of objects presented to it. These cues are then later used to identify examples ofudthe known objects in previously unseen scenes. The work presented in this thesis is basedudon a statistical representation of shape known as a pairwise geometric histogram whichudhas been demonstrated by other researchers in 2-dimensional object recognition tasks. Anudanalysis of the performance of recognition based on this representation has been conductedudand a number of contributions to the original recognition algorithm have been made. Anudimportant property of an object recognition system is its scalability. This is the. abilityudof the system to continue performing as the number of known objects is increased. Theudanalysis of the recognition algorithm presented here considers this issue by relating theudclassification error to the number of stored model objects. An estimate is also made of theudnumber of objects which can be represented uniquely using geometric histograms. One ofudthe main criticisms of the original recognition algorithm based on geometric histogramsudwas the inability to recognise objects at different scales. An algorithm is presented hereudthat is able to recognise objects over a range of scale using the geometric histogramudrepresentation. Finally, a novel pairwise geometric histogram representation for arbitraryudsurfaces has been proposed. This inherits many of the advantages of the 2-dimensionaludshape descriptor but enables recognition of 3-dimensional object from arbitrary viewpoints.
展开▼
