The paper introduces a new image quality assessment (IQA) model called Gradient Magnitude Similarity Deviation (GMSD), which is designed to efficiently evaluate the perceptual quality of images in various applications such as image compression, restoration, and multimedia streaming. GMSD leverages the sensitivity of image gradients to distortions and uses a novel pooling strategy—standard deviation of the pixel-wise gradient magnitude similarity (GMS) map—to predict image quality accurately. The proposed method is significantly faster than most state-of-the-art IQA methods while maintaining high prediction accuracy. The authors demonstrate the effectiveness of GMSD through extensive experiments on three large-scale IQA databases (LIVE, CSIQ, and TID2008) and compare it with 11 other FR-IQA models. GMSD outperforms all competing models in terms of prediction accuracy, monotonicity, and consistency. The paper also discusses the computational complexity of GMSD, showing that it scales linearly with image size, making it suitable for real-time applications. Finally, the authors highlight the potential of GMSD in various image processing applications and the need for new IQA databases to better reflect human perception.The paper introduces a new image quality assessment (IQA) model called Gradient Magnitude Similarity Deviation (GMSD), which is designed to efficiently evaluate the perceptual quality of images in various applications such as image compression, restoration, and multimedia streaming. GMSD leverages the sensitivity of image gradients to distortions and uses a novel pooling strategy—standard deviation of the pixel-wise gradient magnitude similarity (GMS) map—to predict image quality accurately. The proposed method is significantly faster than most state-of-the-art IQA methods while maintaining high prediction accuracy. The authors demonstrate the effectiveness of GMSD through extensive experiments on three large-scale IQA databases (LIVE, CSIQ, and TID2008) and compare it with 11 other FR-IQA models. GMSD outperforms all competing models in terms of prediction accuracy, monotonicity, and consistency. The paper also discusses the computational complexity of GMSD, showing that it scales linearly with image size, making it suitable for real-time applications. Finally, the authors highlight the potential of GMSD in various image processing applications and the need for new IQA databases to better reflect human perception.