Focus and context in mixed reality by modulating first order salient features

Abstract

We present a technique for dynamically directing a viewer's attention to a focus object by analyzing and modulating bottom-up salient features of a video feed. Rather than applying a static modulation strategy, we inspect the original image's saliency map, and modify the image automatically to favor the focus object. Image fragments are adaptively darkened, lightened and manipulated in hue according to local contrast information rather than global parameters. The goal is to suggest rather than force the attention of the user towards a specific location. The technique's goal is to apply only minimal changes to an image, while achieving a desired difference of saliency between focus and context regions of the image. Our technique exhibits temporal and spatial coherence and runs at interactive frame rates using GPU shaders. We present several application examples from the field of Mixed Reality, or more precisely Mediated Reality. © 2010 Springer-Verlag Berlin Heidelberg.