Surfaces in Range Image Understanding

The problem considered in the underlying book may be stated as follows: given a large number of raw (range) image data, how can these be converted into a small number of surface primitives with global properties, without making too much assumptions about specific objects. In other words, the raw signal data of a range image should be converted into intermediate level symbol primitives (smooth surface regions) by a data-driven (not a model-driven) approach, and as such a segmentation of the image will be created. In contrast to the usual segmentation techniques based on detection of discontinuity (edges), surface primitives will be detected by continuity detection; this is plausible, since the number of pixels representing spatial coherence in an image is usually much larger than the number of pixels representing discontinuity. Studying such grouping of pixels based on continuity is also simpler for range images than for intensity images, since the range values themselves depend on scene geometry and not on illumination or reflection. However, the methods given by the author are also applicable for segmenting intensity images having unsignificant texture. After an introductory chapter with an extensive literature review about surface characterization and image segmentation, which are the two central topics of the book, and a descriptive chapter about the analytical framework for object recognition and segmentation, the author gives a thorough description of possible surface characteristics (mean curvature, Gaussian curvature,...) and illustrates them for different images. In chapter 4 the surface curvature sign characteristics are used to compute variable order surface equations that fit the regions. The author gives ample comments about the choice of the approximating functions by considering the shortcomings of different types, and so proposes finally bivariate polynomials up to the fourth oder as the most useful ones for his purposes. The next chapter describes the grouping of surface primitives into a single description whenever this is possible, and in chapter 6 it is shown how the sign-of-curvature function may be used for edge description. In chapter 7 a lot of experimental results, both for range images and intensity images, are presented. The book of P. Besl, a revised version of his Ph. D. dissertation, is written in a lucid style, with ample explanations about why and how of the proposed methods. As the author states in chapter 8 (Conclusions and future directions), more research is needed for good higher-level vision, and he points out some directions in which this may be done. As a whole the book will be wellcomed by all scientists working in and studying image understanding.

Recommendations

Cited in

(14)

This page was built for publication: Surfaces in Range Image Understanding

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4041046)