Over the last two decades, the World Wide Web has grown drastically as a market, with image and text advertising as one of the leading revenue streams. In practice, a web publisher designates a rectangular region, and an advertiser designs an ad to fit that region. This burdens both the publisher, potentially having to waste ad-space, and the advertiser, having to generate different versions of the same ad to fit various dimensions. An ideal solution would accommodate any ad to fit any region, further allowing for the monetization of currently unusable ad-spaces within websites. This work investigates such a solution: the automatic generation of visual banners of any size, given a single prototype ad. Specifically, the contributions of this work are an optimization framework for determining resized ad parameters as well as a novel saliency map algorithm for estimating regions of importance within an image, which outperformed conventional methods by over 20% in a user study.
展开▼