This text is the second of three volumes by Burger and Burge to provide an algorithmic introduction to digital image processing. The first volume [1] concentrates on introductory material; this volume expands on the material by adding implementation details to a number of the presented algorithms. This textbook is for academicians who want a mathematical or theoretical foundation that underlies the methods used, as well as for engineers who are interested in practical implementations of the most important algorithms. The volume is a very nice extension to the material and a worthwhile read.

The introductory chapter covers the basics of programming with images. The second chapter jumps into image analysis by defining and finding image regions; a number of different methods are discussed: flood filling, sequential region labeling, and contour finding. Implementation details of how to represent the image region as a matrix and how to represent the image using run-length encoding (RLE) are provided. Properties of binary regions are developed, including geometric, statistical, and moment-based shape features, as well as topological properties. The book takes a top-down approach and moves from regions (chapter 2) to simple curves (chapter 3) to a special type of curve, corners (chapter 4).

Chapter 3 detects simple curves. The first issue is retrieving salient structures from the image. The classical Hough transform, as improved by Duda and Hart [2], is used as the basis of a voting scheme to determine consistency among pixels with a particular curve interpretation, be it a line, circle, or ellipse. To accomplish this, the implementation utilizes an accumulator array. Chapter 4 uses the Harris and Stephens corner detector [3,4] to detect corners, by computing the corner response function to select good corner points. It should be noted that from an image detection point of view, gestalt vision psychophysicists avoid corner detection because of the variance it has to the viewpoint. More recently, corner detection based on the Harris operator is invariant: “interest points obtained from the multi-scale Harris operator with automatic scale selection are invariant to translations, rotations and uniform rescalings in the spatial domain” [5]. Readers are referred to the work of Lindeberg [6] for a full treatment. Appendix B provides source code for the major algorithms presented so far.

Color theory comprises the discussions in chapter 5, “Color Quantization,” and chapter 6, “Colorimetric Color Spaces.” Chapter 5 presents a number of algorithms to convert the current 24-bit color image to an eight-bit color image, such as Heckbert’s median cut [7], Gervautz and Purgathofer’s octree [8], and the populosity algorithm, which constructs a histogram to count the number of times each color appears and then chooses the most frequent colors. Other methods for vector quantization are also presented. Chapter 6 provides an extensive treatment of various colorimetric spaces (variants of CIE and RGB, as well as an L*a*b* color space implementation).

Chapters 7 to 9 involve more advanced mathematical development and analyze different types of transforms. Chapter 7 introduces spectral techniques with the classic Fourier transform and discusses important properties of the Fourier transform. The rest of the chapter works with discrete signals and implements the discrete Fourier transform (DFT). The treatment is from a teaching perspective, so each subsection addresses a specific aspect of the theory. Building on this theory treatment, chapter 8 implements the two-dimensional (2D) DFT. Many aspects of this transform are discussed, including windowing and the effects of periodicity. Chapter 9 differentiates the discrete cosine transform (DCT) by one and two dimensionality and the notions of separability.

The last two chapters are quite interesting. Chapter 10, “Geometric Operations,” concisely presents all types of mapping and interpolation techniques, and Java implementation details are discussed. Unlike previous chapters that focus on image analysis, this chapter is typically used for object reconstruction and the display of images. The last chapter has a higher-level vision application in mind, comparing images as a whole and for patterns within.