Catálogo de publicaciones - libros

Imagine the following scenario: you are visiting the street carnival in Cologne, Germany for the first time. Fascinated by the colorful and imaginative costumes of the people around you, your gaze wanders from one exciting spot to the next: here a clown with a fancy dress, there a small boy masqueraded as Harry Potter. But not only visual cues capture your attention: over there a band starts to play the new hit of the year and the smell of fresh cookies from the right also revives your interest.

Visual attention is, as mentioned in the introduction, the selective process that enables us to act effectively in our complex environment. The term attention is common in everyday language and familiar to everyone. Nevertheless - or even therefore - it is necessary to clarify and define the term properly. Since visual attention is a concept of human perception, it is important to understand the underlying visual processing in the brain and to know about the psychophysical and neuro-biological findings in this field.

The increased interest on research on visual attention together with the increased power of computers and the resulting ability to realize complex computer vision systems has led to a wide variety of computational systems on visual attention. In this chapter, we will review the most influential work in this field. We already considered models of visual attention in the previous chapter. Although several of them are also implemented computationally, their focus is on the psychological aspect of visual attention more than on the technical aspect: the models of the previous chapter try to explain and better understand human perception whereas the systems in this chapter usually have the aim to improve vision systems for applications in computer vision and robotics.

In the previous chapter, we introduced several computational attention systems of the current state of the art and discussed their characteristics and limitations. In this and the following chapters, we present the new visual attention system VOCUS (Visual Object detection with a CompUtational attention System) which extends and outperforms the current approaches in several aspects, yielding an innovative, efficient, and robust system for detecting regions of interest. We start by introducing the bottom-up part of VOCUS that detects saliencies based merely on the image data in this chapter before we consider top-down influences in chapter 5.

Detecting regions of interest with visual attention is an important mechanism in human visual perception. However, what is of interest depends on the situation. In the previous chapter, we focused on simulating bottom-up mechanisms of visual attention. These define regions as interesting which have a high contrast to their surroundings and are unique in the setting. As mentioned in chapter 2, top-down influences also play an important role in human visual attention: knowledge, motivations, emotions, and goals define what is of interest in a certain situation.

In the previous chapters, we have dealt exclusively with the part of attention that is concerned with visual processing. This part is the best investigated one in human behavior, probably because vision is the sense using the most capacity in the human brain: the 32 representations of the retina occupy more than half of the whole cortex [Kandel et al., 1996] and the primary visual cortex V1 has the richest architecture of all cortical areas [Zeki, 1993].

According to [Neisser, 1967], object recognition in human perception is done in two steps: first, attentional processes select a region of interest, and second, complex object recognition is restricted to these regions. In the previous chapters, we introduced the computational attention system VOCUS that performs the first of these steps. In this chapter, we realize the second step: VOCUS is combined with a well-known classifier [Viola and Jones, 2004] resulting in a complete recognition system. This approach is called attentive classification (cf. Fig. 7.1).

In this thesis, we have introduced the new computational attention system VOCUS for the efficient and robust detection of regions of interest in images. The approach regards object recognition as a two step process: first, the fast attention system detects regions of interest in the whole image and second, a classifier recognizes the content in the specified region. This separation enables an efficient processing since complex object recognition is restricted to a small image region.

Here, we will describe some of the techniques and methods of computer vision that are used in this work. The description is intended for the reader who is not familiar with computer vision. It is far from being an exhaustive introduction into the field. For further reading please refer to [Forsyth and Ponce, 2003, Gonzales and Woods, 1992, Phillips, 1994]. Note also that many of the presented techniques are provided by the Open Source Computer Vision Library OpenCV [OpenCV, 2004].

Here, we present the details of the classification method of Viola & Jones that was introduced in chapter 7. The classifier was also described in our publications [Frintrop et al., 2004b] and [Mitri et al., 2005]. Further details can be found in the original papers [Lienhart and Maydt, 2002,Viola and Jones, 2004].