基于信息熵和深度学习的无参考图像质量评价方法研究*

现有使用卷积神经网络进行图像质量评价的方法，存在训练数据量不足、局部图像块失真分数不能确定等问题。针对这些问题，本文提出了一种基于信息熵的卷积网络训练方法，在考虑信息熵对图像质量影响的基础上，将LIVE数据集中的失真图像进行分块处理，扩大训练集。同时，用分块后的信息熵代表不同分块对失真图像的质量影响程度，并根据该信息熵对卷积神经网络的损失函数进行了调整。实验结果表明，本文提出的模型预测结果更为接近人类视觉感知。

Research on Non - reference Image Quality Evaluation Method Based on Information Entropy and Deep Learning

In the method of image quality evaluation with the convolution neural network, there are some problems such as insufficient training data and the distortion score of the partial image block. Dealing with these problems, this paper proposes a convolution network training method based on information entropy. Considering the influence of information entropy on image quality, the distorted image in the LIVE data set is fragmented to extend the training set. At the same time, the information entropy are represent the mass impact of different blocks on the quality of the distorted image, adjusting the loss function of convolution neural network. Experimental results show that the proposed model is closer to human visual perception.