基于深度学习模型的图像质量评价方法

Image quality assessment based on deep learning model

为了有效提取与视觉感知质量高度相关的图像特征,改进图像质量评价方法,在深度学习的框架下,提出一个全新的卷积神经网络IQF-CNN结构,能自动学习判别性更强的图像质量特征,并利用学习的特征进行图像质量评价.同时,该算法采用局部亮度系数归一化、dropout等技术进一步提高网络学习能力.实验结果表明:该算法能较准确地评估五种常用的图像失真,尤其在JPEG压缩、JPEG2000压缩和高斯模糊图像失真上与人眼主观感知质量具有很高的一致性,整体性能比较优于其他经典评价方法.

Extracting image features highly correlated with visual perception is important for mage quality assessment .In this paper ,a novel convolution neural network IQF-CNN（convolutional neural network based on image quality features） was proposed .This model learned more discriminative im-age quality features ,and used the learned features for image quality assessment .Meanwhile ,the local luminance coefficients normalization and dropout technology were used to improve the model learning ability .The experimental results show that the algorithm can accurately evaluate the five common im-age distortions ,especially for JPEG ,JPEG2000 and BLUR .The overall performance is better than other classical assessment methods .