Learning Influence of Content on Image Quality

Navaneeth Kamballur Kottayil

In this project we derive a computational strategy to enhance the performance of Image Quality Metrics (IQM) by using content specific features of an image. We do this by creating Visual Error Importance (VEI) map that is applied to the error maps computed by the IQM. A global optimization can be used to compute the VEI map that is optimal for any given IQM from a set of simple image features.

Error importance maps if we artificially force the algorithm to simulate a scene content is shown below.

Note how error importance change; in outdoor natural scenes, VEI emphasizes errors in smooth, well lit areas and lower to medium complexity regions, with a lower priority to salient objects. Such areas are more important if natural scenes are being displayed. On the other hand, for outdoor scenes with man-made structures, we see that VEI assigns more importance, in term of image quality, to regions with higher details, and less importance to smooth regions. For indoor scenes, we observe a larger importance given to smoother regions.

Our results show that VEI maps produce improvement in performance of IQA in all the cases we tested.