Image quality models for extended sets of distortions

The PQS by Miyahara (1988) as well as the popular signal-to-noise ratio (S/N) are tested for a more extended set of picture data. The S/N measure is considered, since it is simple, and obvious in the context of the human observer due to an inclusion of a logarithmic function. An important result is that PQS performance is less robust against variation of distortions. Moreover, it is shown that a PQS modified by introducing a simple nonlinear regression analysis leads to improved performance, particularly for more extended sets of distortions