6
Validation and Evaluation for Visual Attention Models
It is meaningful and necessary to validate and benchmark the computational models (with both bottom-up and top-down mechanisms) described in the previous chapters [1–10] with multiple types of ground-truth and different databases from independent sources [11–14] (i.e., with different visual contents, and by different subjects and laboratories). This chapter introduces the evaluation methods as well as related ground-truth databases. We believe that such cross-source tests also facilitate model improvement/optimization, determination of the scope of applications for a model, and parameterization to enforce better alignment with subjective data.