Abstract
Automatic texture fidelity assessment that would correspond to the human visual perception is an important, but still unsolved computer vision problem with numerous useful applications in the various vision application areas such as image compression and modeling, video streaming or fast image database retrieval. The problem is not satisfyingly solved even for the most simple static monospectral texture representation thus progress in the automatic assessment of texture fidelity is required.
We propose improved multiresolution texture fidelity measure based on Markovian random field texture model, which correlates well with human texture fidelity evaluation obtained from texture fidelity benchmark.