Beyond pixel
Jia Li · 2015
This paper will discuss the problem of visual saliency estimation from the perspective of saliency context. Existing saliency models, which may have remarkable differences in saliency mechanisms and computational frameworks, differ mainly in two aspects: which kinds of saliency contexts are adopted and how to assess saliency based on the selected contexts. In other words, context defines everything in saliency. By analyzing recent trends in selecting saliency context, we find that the usage of adaptive context could be a feasible direction toward better saliency models, which we will explore in our feature research efforts.