Effective preprocessing stage in the fourier transform domain for image quality assessment

Image quality assessment (IQA) is currently an important research topic. In fact, the varying viewing distance between audiences and the display seriously affect the IQA accuracy, which has been largely overlooked. To this end, in this paper we take into account the image size and viewing distance as well as the preferential activation of V1 cells by vertical and horizontal contours, and thereby propose an adaptive frequency selection (AFS) model to preprocess input visual signals before IQA metrics are used. Our algorithm works by first applying Fourier Transform (FT) to the reference and distorted images to approximate the low-pass behavior of the human visual system, then extracting proper amount of low frequency components and partial high frequency components corresponding to vertical and horizontal directions, and finally reconstructing spatial images with the inverse FT. We validate the performance of AFS based PSNR and SSIM on the related LIVE, Toyama and IVC databases with clearly specified viewing conditions. Experimental results and comparative studies show the effectiveness of the proposed model.