Real Time Dynamic Image Re-targeting Based on a Dynamic Visual Attention Model

With the constantly growing number of multimedia devices, images can now be viewed on a broad range of display of tremendously varying size. However, perceiving all the details of a multimedia content is difficult on smaller mobile devices. To solve this problem, saliency based image re-targeting algorithms have been proposed in order to generate more perceptually efficient thumbnails. In this article, we propose a new real-time visual attention based image re-targeting method which differs from previous works in that it generates dynamic (i.e. time changing) thumbnails. This unique behavior is made possible by the use of a dynamic computational model of visual attention. By exploiting the singular characteristics of this model, we are able to dynamically re-target large and complex images in order to efficiently script their discovery on large and small displays. We achieved such re-targeting by integrating simulated attentional focuses over a fixed time window into a temporal heat-map. By segmenting this heat-map at each time step, we can generate the region of interest that will be displayed as dynamic preview. In order to validate that every salient location of the image has been displayed, we compare the regions shown in the preview with mean observers eye fixations obtained during a free viewing experiment.