No-reference Image Quality Assessment Using Shearlet Transform

Image and video quality measurements are crucial for many applications, such asacquisition, compression, transmission, enhancement, and reproduction. Nowadays,no-reference (NR) image quality assessment (QA) methods have been drawn extensiveattention because it does not need any information of reference images. However, mostproposed no-reference quality methods are designed only for one or a set of predefinedspecific distortion types, which are unlikely to generalize for evaluating images distortedwith other types of distortions.In order to assess different kind of image distortions, in this paper, a novel NR QAmethod is proposed which is based on Shearlet transform. As the third generation wavelet,Shearlets have their own superiority. Anisotropic dilation operator and shear operator areintroduced to overcome the shortcomings of traditional wavelets. Thus, this newmultiscale directional transform has a strong ability to localize distributed discontinuities.In this paper, we first analyze three methods to implement the digital Shearlet transform,that is, the Shearlet transform in spatial domain, Fourier based Shearlet transform and theShearlet transform based on pseudo-polar FFT. Then, we analyze the statistical differencebetween natural images and distorted images. The distorted image leads to significantvariation in the distributed discontinuities in all directions. Thus, the statistical property ofthe distorted image is significantly different from that of natural images in Shearletdomain. A new model is also proposed to measure this difference.Numerical experiments demonstrate that this new NR QA method is consistent withsubjective assessment, very effective for many well-known types of image distortions andsuperior to the existing prominent methods.