Measuring digital image quality is of importance in numerous applied research fields, such as image acquisition, processing or reconstruction. The choice of quality measure directly affects the outcome of evaluation, e.g. when assessing newly proposed algorithms, or even controls the underlying process itself, e.g. when using loss functions for neural network learning. Subjective evaluation, i.e. assessment by human beings, is often too time-consuming and expensive. Moreover, manual annotations tend to suffer from inconsistencies, especially in complicated medical problems. To overcome these issues automated image quality assessment (IQA) is needed, yielding fast and consistent evaluation schemes. Gabor representations and their generalizations are a common choice for analyzing audio and other 1-dim signals. Applying discrete Gabor transforms in two or higher dimensions has received less attention, although for some problems in image processing they are useful and rather straight-forward to apply and interpret. Since Gabor filters are known to be closely related to mammalian brains and to represent texture well, they are an obvious choice for IQA. In this talk we will discuss the influence of different image representations in the context of IQA. In particular, by conducting numerical experiments we will demonstrate the impact of parameter choices in Gabor filters on the resulting IQA measure.

https://univienna.zoom.us/j/66031419470?pwd=bXd3V0xEMWM0MTQwS09nWStEV0NnUT09