Blind Noisy Image Quality Assessment Using Variance, Kurtosis, and Noise Estimation Score
Ting He, Yizhi Liu, Youya Xu · 2021 3rd International Academic Exchange Conference on Science and Technology Innovation (IAECST) · 2021
To meet image quality requirements in daily applications and the evaluation of image algorithm performance, the image quality assessment has been widely researched. The discrete cosine transform (DCT) does not effectively represent mutations or singularities, and the AWGN usually appears in the DWT sub-bands at high frequencies, so DWT has a better choice than DCT to respond to distinct high frequencies. Based on the above characteristics, the proposed algorithm extracted variance, kurtosis, and noise estimation score features to assess image quality, and trained the above features by SVR model, and proposed an improved method to assess the quality of the image. The proposed algorithm was simulated in the CSIQ, LIVE, and TID database. Then, the three databases were synthesized together to a new database and tested algorithm performance for improving generality. The results show that the improved model has better evaluation results compared to the existing no-reference noise evaluation model.