1.5 Scope of This Book

During the past decades, research on visual selective attention has received a great deal of interest from both biologists and engineers. Computational models of visual attention are being touted as an exciting new methodology for computer vision and image processing due to the apparent ability to imitate human vision and overcome the challenging issues of computer vision. The content described in this book has been prepared for most students and professionals who are entering or already working in the related areas. We aim to help readers to understand the issues relevant to visual attention in a systematic way, and to be able to select or build an attention model suitable for their applications based upon the knowledge and insight delivered in this book. In addition, we also provides many examples and case studies for scientists, engineers and students on how to solve various problems based upon scientific first principles and practical requirements.

A proper understanding of the biological aspects of visual attention requires knowledge in a variety of fields including the structure of the biological visual system in anatomy, the important relevant conclusions in psychology and physiology, signal and information theory and artificial neural networks. Thus, Part I of this book presents the necessary introduction to the concepts of attention and related knowledge, so that beginners without prior related knowledge can learn the necessary background. Part II of this book describes some typical computational visual attention models according to the concepts and theory presented in Part I, and gives methods of how to compare and test different computational models; a few software websites and code of MATLAB® for existing computational models are listed in the references of Part II or provided by the book, for the convenience of the reader. The applications of computational models for visual attention are presented in Part III. Finally this book discusses several controversial issues in visual attention and the possible future work, based on our experience and understanding of visual attention modelling and applications (in both academic research and industrial development).

There are two chapters in Part I. In Chapter 1 (this chapter), we have presented a concise and interesting introduction of visual attention phenomena and the challenges of computational modelling. Some basic concepts and types of visual attention are defined. We have reviewed three phases of visual attention model development: biological studies, computational modelling and applications. Chapter 2 firstly introduces the structure and properties of the human visual system (HVS) in anatomy, as well as the latest knowledge and findings on visual attention selection in psychological and cognitive science. The main theories in visual attention models are then introduced, such as feature integration, synchronized oscillation, competition and inhibition between cells, guided search theory, redundancy reduction and information theory. This background knowledge forms the basis for building computational models in practice.

Part II has four chapters, Chapters 03, 04, 05, 06. Chapter 3 presents bottom-up based computational models in the spatial domain. Itti et al.'s model of 1998 is described in detail. Other models based on Itti et al.'s model also are discussed. Then the models based on information theory, such as AIM, Bayesian approach and SUN, are presented. In Chapter 4 bottom-up based models in the frequency domain are introduced and discussed. Frequency domain models have fast speed to meet real-time requirements, so they can be easily implemented for engineering tasks. Also, the biological justification for frequency models is provided mathematically and biologically. Chapter 5 then gives typical attentive models combining both top-down and bottom-up attention. The relevant models are presented in the chapter, and the top-down knowledge and working memory from neural network learning and the decision tree algorithm are also discussed. In Chapter 6, methods for testing and validation of built models against ground-truth data (including eye tracking results) are discussed. This helps in building new computational attention models and it benchmarks them with other existing models.

In Part III, we allocate two chapters to explain how to use computational models to apply to computer vision and image processing. Chapter 7 presents object detection and recognition by using bottom-up and top-down models. Some application examples for robot navigation and object recognition show that visual attention can indeed solve some of the challenges of computer vision. Then attention based image retrieval is also described. Chapter 8 covers applications in image and video processing and quality assessment. The traditional quality assessment metrics (such as peak signal-to-noise ratio) in image and video compression and processing do not align with human subjective perception, so the application of attention models and the other relevant properties in the HVS (e.g., just-noticeable difference) improves the process of visual quality assessment. Other image processing tasks using vision attention models such as image resizing and compressive sampling are also considered in Chapter 8.

The final chapter is a summary of this book and conclusions. Several controversial issues in visual attention and its modelling are discussed, and these indicate some potential future work in the related studies.