IMAGE GRADIENT SEGMENTATION (SHADING, PROCESSING, ANALYSIS)

In this thesis we describe an approach to identification of shaded segments in an image providing a model for the curved surface that generated the segment. The analysis, development, implementation, and evaluation of an algorithm that obtains the segments are explained. The output of the algorithm is a list of the image segments where shading was identified along with the geometric and curved surface model parameters of each segment.;The Image Gradient Segmentation Algorithm includes three phases: (1) Partition of the BxBy map into components with curved surface models, (2) Segmentation of the image into connected regions corresponding to the BxBy map partitions, and (3) Identification of curved surface models. The algorithm has been implemented in a stand-alone microprocessor-based system. Examples of Image Gradient Segmentation are provided for synthesized as well as real things.;The algorithm provides a description of images in terms of components which correspond to curved surfaces in scenes. Examples described here demonstrate the feasibility and reliability of the algorithm. Future work will consider the integration of this image description into systems which interpret more general scenes.;The Image Gradient (BxBy) map, a two-dimensional plot of the distribution of image points in the image gradient space, is utilized as a tool in the development of the algorithm. Properties of the surface gradient and image gradient maps for curved surface models are analyzed. The BxBy map is affected by noise in the image, and detailed analysis of the effects of noise in the image gradient space is presented. The noise analysis is used in the development of the algorithm.