Blind image quality assessment based on statistical features.

No-Reference Image Quality Assessment (NR-IQA) is challenging due to the absence of reference images. Natural Scene Statistics (NSS) provides a robust foundation for NR-IQA. However, BRISQUE relies on the statistical features of image intensity and imposes a strong zero-mean assumption, which may not hold in practice and could limit its performance. This paper proposes a novel NR-IQA method, the Blind Image Quality Assessment Based on Statistical Features (BIQABSF) algorithm, which introduces two key improvements. First, we utilize the statistical features of log-contrast (LC), which more closely align with human visual perception. Second, the distribution of the locally normalized LC was modeled using a generalized Gaussian distribution (GGD) without a zero-mean constraint, based on the empirical observation of its asymmetric nature. Specifically, we compute the Local Mean Subtracted Log-Contrast (LMSLC) and extract its distribution parameters as features for training a Support Vector Regression (SVR) model. Extensive experiments on the LIVE, TID2013,KADID-10k, and CSIQ databases demonstrated that BIQABSF achieved highly competitive performance, even surpassing several deep learning-based methods in terms of efficiency and generalization, thereby demonstrating its potential as an efficient and generalizable solution for real-world applications.