Determining the exact purpose of this short book is not easy. Among its eight chapters, the first three, “Introduction,” “Segmentation and Vectorization,” and “Text and Graphics Analysis in Engineering Drawings,” have the appearance of survey papers, with many bibliographic references but only shallow discussions. The fourth chapter, “A Structural Model for Engineering Drawings,” is in fact the master’s thesis of one of the author’s students. Chapter 5, “Non-Deterministic Analysis Systems (NDAS),” is also by this student and is also based on her master’s thesis.

After a very short sixth chapter that introduces the idea of map analysis, chapter 7, ”Road and Road Intersection Extraction,” is by another student of the author, and based on his master’s thesis. The final chapter, ”Other Semantic Feature Segmentation,” is a modified version of a previous paper presented in a workshop, and one of the three coauthors is the student who wrote chapter 7.

Thus, chapters 4, 5, and 7, which form the bulk of the book (120 of the book’s 200 pages), are authored by people not even mentioned on the book’s cover, and chapter 8 (20 pages) is only partly authored by the book’s author, though the degree of the author’s involvement is not described. These chapters are not really research papers since they only describe a specific method without comparison to other similar work.

The idea of the book is to present a particular method for the automatic analysis of engineering drawings and road maps. I cannot say why the qualifier “raster“ in the book’s title is appended only to maps--the drawings are also raster images. In the examples, the images seem to be of a rather poor quality, although the originals should be available.

Engineering drawings mainly consist of straight lines delimiting parts of the drawings, explanatory text, and simple graphics like sketches of mechanical parts with dimensions. The tool used, MATLAB nondeterministic agent system (MNDAS), is programmed in MATLAB and uses a system of independent agents for extracting low-level image features and higher-level annotation structures.

Maps are more complicated and need a lot more semantic analysis for segmenting the map components and characterizing them between, for example, background, vegetation, roads, water, political lines, and iso-contours. This obviously necessitates having some previous knowledge about the contents of the documents to be analyzed.

This is an interesting subject, but it’s impossible to know whether the method proposed by the author is better than others, or more successful, or less costly. Although Appendix B contains a very short MNDAS user manual, this is clearly not enough for knowing how to use it, or even for knowing whether it is available to others.