No-reference image quality assessment using statistical wavelet-packet features

• We present a no-reference image quality assessment method in the wavelet-packet domain.
• The proposed method achieves a high accuracy as compared to existing methods.
• The method is able to assess image quality across various distortion types.
• The method is general-purpose and it is designed in a modular manner.

In this paper an efficient no-reference (NR) image quality assessment (IQA) method is presented based on the statistical features of subband coefficients in the wavelet-packet domain. The proposed method is based on the hypothesis that potential distortions may alter the statistical characteristics of natural un-distorted images. Hence, by characterizing the statistical properties of a given distorted image one can identify the distortion and its strength in the distorted image. For this purpose, several statistical features of a given gray-scale image as well as the magnitude of its gradient and its Laplacian are extracted in the wavelet-packet domain. The extracted features are then mapped to quality scores within a two-stage quality assessment framework. The proposed method is general-purpose, and is able to assess the image quality across various distortion categories. Experimental results indicate that the proposed method achieves high accuracy in image quality prediction as compared to several prominent and state-of-the-art full-reference and no-reference IQA methods.