Blind Image Quality Assessment Model Based on Deep Convolutional Neural Network

As learning-based models become predominant in image processing field, especially in image classification and object detection, image quality assessment (IQA) has been a hot issue since the quality of train images is of great importance fro model performance. This paper proposes a novel blind IQA model evolved from VGG-16 [1], a state-of-the-art classification deep convolutional neural network(CNN). It modifies the classification network to apply it to regression task by keeping only one neuron in the last fully-connected layer, the output of which is considered as the quality assessment metric after being normalized to a score between 0 and 1 via Sigmoid function. And experimental results demonstrate that this model achieves a superb assessment performance on TID-2013 dataset [2].