The paper introduces a novel approach, NIMA (Neural Image Assessment), for predicting both technical and aesthetic qualities of images using deep convolutional neural networks (CNNs). Unlike existing methods that typically predict the mean opinion score, NIMA predicts the distribution of human opinion scores, which provides a more accurate and correlated prediction with human perception. The method leverages state-of-the-art deep object recognition networks, such as VGG16, Inception-v2, and MobileNet, and uses the squared Earth Mover's Distance (EMD) loss to train the models. The proposed approach is evaluated on three datasets: AVA, TID2013, and LIVE, showing superior performance in predicting mean scores and standard deviations. Additionally, the method is applied to image enhancement tasks, such as tone adjustment and denoising, demonstrating its ability to guide these processes to produce perceptually superior results. The computational complexity of the models is also discussed, with MobileNet being the most efficient. Overall, NIMA offers a robust and accurate solution for image quality assessment and enhancement.The paper introduces a novel approach, NIMA (Neural Image Assessment), for predicting both technical and aesthetic qualities of images using deep convolutional neural networks (CNNs). Unlike existing methods that typically predict the mean opinion score, NIMA predicts the distribution of human opinion scores, which provides a more accurate and correlated prediction with human perception. The method leverages state-of-the-art deep object recognition networks, such as VGG16, Inception-v2, and MobileNet, and uses the squared Earth Mover's Distance (EMD) loss to train the models. The proposed approach is evaluated on three datasets: AVA, TID2013, and LIVE, showing superior performance in predicting mean scores and standard deviations. Additionally, the method is applied to image enhancement tasks, such as tone adjustment and denoising, demonstrating its ability to guide these processes to produce perceptually superior results. The computational complexity of the models is also discussed, with MobileNet being the most efficient. Overall, NIMA offers a robust and accurate solution for image quality assessment and enhancement.