An agent system for advertisement inclusion using human-based computation

In this paper, we will present a system based on intelligent agents, which uses human-based computation (HBC) for advertisement adjustment in images. This system will learn from human interaction to obtain a salience map of the most important parts of an image and will use this information to fit a fixed-sized advertisement in the least important part of the image. Although this approximation has been developed to be used in many different applications, its advantages are more evident when used to add advertisements to real time channels, such as websites or video streaming. In this way, media rich applications (principally online) are the most appropriate for this process. Several studies have demonstrated that badly placed advertisements are frequently ignored by users (and may even provoke irritation) and thus do not serve their purpose. In this way, correct advertisement placement is fundamental to maximize an advertisement's effectiveness. As we will see, our agent system is more robust than previous approximations because it is less influenced by specific image features and takes into account the most important parts of an image from the human point of view. In addition, we will compare our approximation with a classical biological model for visual saliency.

► We use human-based computation (HBC) saliency maps for advertisement inclusion in online media.
► HBC saliency maps detect important features for humans not always reflected in classical saliency maps.
► Large advertisements (>30%>30% of the image) placed using HBC saliency maps are well-fitted and do not disturb image perception, maximizing their power of attraction.
► Classical saliency maps are not well suited for including large advertisements.