The Multiple Window Parameter Transform

Abstract

This paper presents the multiwindow transform, which is an extension of parameter transform techniques that increases performance and scope. This is achieved by exploiting the long-range correlated information contained in multiple portions of an image. Traditional local parameter transforms, which are used for the detection and reconstruction of features, have been successfully employed to find, for example, lines and circles in edge maps. However, as shown by many examples in the literature, there is a steep tradeoff between accuracy and computational complexity when dealing with more complicated (i.e., high-dimensional) geometric features. Multiple window transforms allow for the extraction of high-dimensional features with improvement in accuracy over conventional techniques while keeping linear to low-order polynomial computational and space requirements with respect to image size and dimensionality of the features. Using correlated information provides a direct link between extracted features and supporting regions in the image. This, coupled with evidence integration techniques, is used to suppress noisy or nonexistent feature hypotheses. Parameter spaces are implemented as constraint satisfaction networks, where feature hypotheses with overlapping support in the image compete. After an iterative relaxation phase, surviving hypotheses have disjoint support, forming a segmentation of the image. Examples that show the performance and provide insight in the behavior are given. Complex, high-dimensional, parametric features such as surfaces and surface-intersection curves can be reconstructed from range data. © 1992 IEEE