Relating spatial and spectral models of oriented bandpass natural images
Zeina Sinno, Alan Conrad Bovik · 2016
Univariate models of the Natural Scene Statistics (NSS) of perceived digital pictures have been deployed in a wide variety of image and video processing applications. However, much less effort has been made towards understanding, modeling, and using bivariate image NSS. Towards filling this gap, Su et al. developed a closed form correlation model of oriented bandpass natural images applicable to adjacent pixels. We later extended this model to account for pixels separated by larger spatial distance. Here, we expand our previous work further to model the bivariate responses of bandpass spatial filters covering a wider range of bandwidths. Furthermore, we study the relationship between the parameters of the closed form correlation model and image spectral models.