| dc.description.abstract | This thesis is concerned with the identification of features within two-dimensional imagery. Current acquisition technology is capable of producing very high-resolution images at large frame rates and generating an enormous amount of raw data. Exceeding present signal processing technology in all but the simplest image processing tasks, the visual information contained in these image sequences is tremendous in both spatial and temporal content. A majority of this detail is relatively unimportant for the identification of an object, however, and the motivations for this thesis, at the core, are the study and development of methods that are capable of identifying image features in a highly robust and efficient manor. Biological vision systems have developed methods for coping with high-resolution imagery, and these systems serve as a starting point for designing robust and efficient algorithms capable of identifying features within image sequences. By foveating towards a region of interest, biological systems initially search coarse-scale scene representations and exploit this information to efficiently process finer resolution data. This search procedure is facilitated by the nonlinear distribution of visual sensors within a biological vision system, and the result is a very efficient and robust method for identifying objects. Humans will initially identify peripheral objects as potential regions of interest, acquiring higher-resolution image information by focusing on the region, and deciding if the perceived object is actually present through the use of all available knowledge of the scene. | |
