Shape information distribution and object identification using the generalized Hough transform

A general technique, based on the Hough transform method, to identify objects in a scene from image data has been investigated. The scene contains a number of objects that may partially occlude each other in the image. Techniques for both intensity (2-D) and range imagery have been considered. The conventional Generalized Hough Transform (2-D GHT) is suitable for identifying 2-D boundaries of partially occluded objects. This thesis contains analysis and refinements of the 2-D GHT and extends this transform to range imagery with the development of the 3-D Generalized Hough Transform (3-D GHT).; A refinement of the 2-D GHT, called the Tuned Generalized Hough Transform, has been developed that overcomes much of the noise sensitivity problem of the conventional 2-D GHT. This technique also determines the orientations of the detected object.; In order to select salient shape information in a range image, sampling theorems for slope-limited curves and surfaces have been developed. Two new parameterizations of curves and surfaces are introduced: the shape density identifies points of high shape interest and the shape function may be used to characterize a total curve or surface.; A new version of the Hough transform, called the 3-D Generalized Hough Transform (3-D GHT), has been developed. This transform provides an effective technique for the identification and location determination of 3-D objects. It is suitable for images with multiple objects and occlusion; the surfaces of the objects of interest may have arbitrary complexity. Experimental results indicate that the 3-D GHT has a better performance than the 2-D GHT and can distinguish between some objects that appear identical to the 2-D GHT.; A model to predict the performance of the Generalized Hough Transform (GHT) is presented which includes the effects of additive noise, quantization noise, and the coincident degree of secondary objects in an image. This model verifies the superior performance of the 3-D GHT compared to the 2-D GHT.; A Generalized Projection Transform (GPT) is defined, which not only offers a uniform representation for projection transforms, such as the Radon transform, and the Hough transform and its variations, but also demonstrates the different functions of these two important transforms.