# Concise Computer Vision: An Introduction into Theory and Algorithms (Undergraduate Topics in Computer Science)

## Reinhard Klette

Language: English

Pages: 441

ISBN: 1447163192

Format: PDF / Kindle (mobi) / ePub

Many textbooks on computer vision can be unwieldy and intimidating in their coverage of this extensive discipline. This textbook addresses the need for a concise overview of the fundamentals of this field.

Concise Computer Vision provides an accessible general introduction to the essential topics in computer vision, highlighting the role of important algorithms and mathematical concepts. Classroom-tested programming exercises and review questions are also supplied at the end of each chapter.

Topics and features:

* Provides an introduction to the basic notation and mathematical concepts for describing an image, and the key concepts for mapping an image into an image

* Explains the topologic and geometric basics for analysing image regions and distributions of image values, and discusses identifying patterns in an image

* Introduces optic flow for representing dense motion, and such topics in sparse motion analysis as keypoint detection and descriptor definition, and feature tracking using the Kalman filter

* Describes special approaches for image binarization and segmentation of still images or video frames

* Examines the three basic components of a computer vision system, namely camera geometry and photometry, coordinate systems, and camera calibration

* Reviews different techniques for vision-based 3D shape reconstruction, including the use of structured lighting, stereo vision, and shading-based shape understanding

* Includes a discussion of stereo matchers, and the phase-congruency model for image features

* Presents an introduction into classification and learning, with a detailed description of basic AdaBoost and the use of random forests

This concise and easy to read textbook/reference is ideal for an introductory course at third- or fourth-year level in an undergraduate computer science or engineering programme.

The epipolar profile on the right illustrates that there is a kind of a “post” in the scene at the grid point (7,10) causing that the scene does not satisfy the ordering constraint in row y illustrated by this diagram. Insert 8.3 (Origin of Epipolar Profiles) [Y. Ohta and T. Kanade. Stereo by intra- and inter-scanline search using dynamic programming. IEEE Trans. Pattern Analysis Machine Intelligence, vol. 7, pp. 139–154, 1985] is the paper that introduced epipolar profiles and also pioneered

Fourier basis functions: (1.32) Ignoring the DC component that has the phase zero (and is of no importance for locating an edge), the resulting Fourier transform J is composed of n=(2k+1)2−1 complex numbers z h , each defined by the amplitude r h =∥z h ∥2 and the phase α h , for 1≤h≤n. Figure 1.23 illustrates an addition of four complex numbers represented by the amplitudes and phases, resulting in a complex number z. The four complex numbers (r h ,α h ) are roughly in phase, meaning that the

been essential for establishing random decision forests as a technique for ensemble learning; see, for example, his paper [L. Breiman. Random forests. Machine Learning, vol. 45, pp. 5–32, 2001]. The technique has its roots in tree construction methods and related classifiers, with related pioneering publications by various authors, dating back to about the early 1980s. 10.3.3 Training a Forest We apply supervised learning for generating a random decision forest. We have a set S of samples, i.e.

function g new has the same mean and variance as the function f. Distance Between Two Functions Now we define the distance between two real-valued functions defined on the same discrete domain, say 1,2,…,T: (1.13) (1.14) Both distances are metrics thus satisfying the following axioms of a metric: 1. f=g iff d(f,g)=0, 2. d(f,g)=d(g,f) (symmetry), and 3. d(f,g)≤d(f,h)+d(h,g) for a third function h (triangular inequality). Structural Similarity of Data Measures Assume two different spatial

by σ. The angle between the vector from O to P=(X,Y,0) and the X-axis is called the tilt and denoted by θ. The point P is in the distance sinσ to O. The unit normal n ∘ defines a point on the surface of the Gaussian sphere that is uniquely represented by (σ,θ), also called the point’s spherical coordinates. Fig. 7.6The Gaussian sphere is defined by radius 1 Gradient Space We define an ab coordinate space (see Fig. 7.7, right) where each point (a,b) represents a gradient (a,b)⊤ in the XYZ