DCT-Based No-Reference Quality Assessment of AWGN Images

Images captured by low-end cameras under low-light conditions often suffer from the presence of additive white Gaussian noise (AWGN), resulting in distortion. This paper proposes a novel no-reference quality assessment method for images contaminated with AWGN. The method leverages the dominance of noise in the higher frequency AC coefficients of a discrete cosine transformed (DCT) noisy image. It specifically focuses on the 0.5% highest-frequency DCT coefficients, whose magnitudes increase with increase in noise levels. The average of these coefficients is used to estimate the quality of an AWGN image. Extensive simulations were conducted on the LIVE, CSIQ, and TID2013 databases, resulting in Spearman’s rank-order correlation coefficients of 0.9853, 0.9234, and 0.9078, respectively. These results demonstrate the high accuracy of the proposed metric compared to existing algorithms. Additionally, the proposed method exhibits lower computational complexity, further highlighting its practical advantages. This no-reference quality assessment method addresses the challenges posed by AWGN in low-light images captured by low-end cameras. Its accuracy, efficiency, and applicability make it a valuable tool for various low complexity image processing applications.