An Improved BRISQUE Algorithm Based on Non-zero Mean Generalized Gaussian Model and Global Structural Correlation Coefficients

To solve the problem of the limited description ability of the quality-aware features used in blind/referenceless image spatial quality evaluator (BRISQUE) algorithm and enhance the accuracy and robustness of the BRISQUE algorithm, an improved BRISUQE (IBRISQUE) algorithm was proposed in this paper. First, we adopted non-zero mean symmetric generalized Gaussian distribution (GGD) model to obtain the quality-aware features from mean subtracted and contrast normalized (MSCN) coefficients. Then, we used non-zero mean asymmetric generalized Gaussian distribution (AGGD) to extract the quality-aware features that could represent the local structural distortions from the neighboring MSCN coefficients along four orientations. Finally, we utilized Pearson linear correlation coefficients (PLCC) of neighboring MSCN coefficients from horizontal, vertical, main-diagonal, and secondary-diagonal directions as the quality-aware features reflecting global structural distortions. The IBRISQUE algorithm was tested on the LIVE and TID2013 benchmark databases. Comparing with state-of-the-art image quality assessment algorithms, IBRISQUE algorithm achieves higher prediction accuracy while the efficiency maintains a roughly equal level compared with the original BRISQUE algorithm. The proposed algorithm that strikes a favorable balance between accuracy and complexity outperforms other competing algorithms significantly. © 2018, Beijing China Science Journal Publishing Co. Ltd. All right reserved.