6.1 Simple Man-made Visual Patterns

Simple visual patterns can be made and adopted in various visual attention experiments to explore the visual attention phenomena or validate visual attention models [3, 4, 12, 14]. Studies show that locations with different low-level salient features from surroundings can be detected rapidly and accurately [12, 13]. Normally, a simple visual pattern includes one single target with an obviously different low-level feature (such as colour, intensity, orientation, shape or density) from others in the same visual scene. These patterns often come from psychophysical experiments. Figure 1.4 has already included several examples of man-made visual patterns. Figure 6.1 shows three further such sample patterns [14] and the associated saliency maps [4]. In Figure 6.1(a), the first sample pattern shows one vertical bar among many largely horizontal ones; the target of the second pattern is one circular object surrounded by horizontal bars; and the third sample pattern is formed by one denser cluster of horizontal bars with other bars. As we know, humans will definitely focus on the vertical target, the circular target and the dense cluster in these three patterns, respectively.

Figure 6.1 Simple visual patterns and their saliency maps. (a) Reproduced with permission from Christopher G. Healey, ‘Perception in Visualization’, North Carolina State University, http://www.csc.ncsu.edu/faculty/healey/PP/index.html (accessed 1 October 2012). (b) © 2008 IEEE. Reprinted, with permission, from C. Guo, Q. Ma, L. Zhang, “Spatio-temporal Saliency detection using phase spectrum of quaternion Fourier transform”, IEEE Conference on Computer Vision and Pattern Recognition, June 2008

Therefore, one evaluation method for visual attention models is to check whether the model under consideration can detect each of the salient targets that psychophysical experiments have authenticated [3, 4] for the patterns exemplified in Figure 6.1(a). Figure 6.1(b) illustrates the saliency maps derived from the visual attention model in [4] for the three patterns, and demonstrates that the targets are detected successfully and convincingly by the model used. A database including more patterns of a similar nature can be found in [14].

Simple man-made patterns have the advantage of clear meaning and non-ambiguity in their design, and the well-acknowledged psychological findings. Therefore, the test results are able to explicitly suggest the exact strengths and weaknesses of a model (e.g., the performance with a particular feature, such as orientation), and possible directions and aspects for model improvement. However, the man-made patterns have a drawback: their use is not general and is insufficient for a practical model since they merely contain a single visual stimulus (i.e., the deviated real target pattern) at a time and the context of the visual stimuli is too simple. In contrast, a real-world visual signal includes many stimuli appearing simultaneously and the context is usually complex. Therefore, it would be better to further test a model with other types of ground-truth data as discussed next.