Abstract
To support pan-contextual access to complex scene images, a pattern sensitive re-sampling scheme is introduced. By restricting ‘what-to-be-observed’ in future scenes to really existing objects exhibiting connected distributions of visual saliency, in this scheme, chromatic diversity and brightness distribution are jointly identified with invariant measure to be associated with fractal attractors. Through infinitely divisible representation of capturing probability, the image features are interpolated as mathematical expectation of infinite resolution. The re-sampling scheme has been verified by using naturally complex scene images.
