摘要
随着立体图像的广泛应用,迫切需要一个具有通用性的工具来评估立体图像的视觉质量,因此提出一种基于卷积神经网络的无参考立体图像质量评价算法。首先使用平面图像数据集对算法框架的主体结构质量图生成网络进行训练;然后使用训练好的网络预测立体图像融合视点图像的质量;最后使用一种加权融合方法得到最终的立体图像质量分数。试验结果表明,算法框架具有相对较好的准确性和鲁棒性。
With the prevailing application of stereoscopic images,a universal tool is urgently needed to evaluate the visual quality of those images.Therefore,a non-reference stereoscopic image quality evaluation algorithm was proposed on the basis of convolutional neural network.Firstly,the planar image dataset was used to train the generation network of main structure quality map of the algorithm framework.Then,the well-trained network was used to predict the quality of the fusion viewpoint images.Finally,a weighted fusion method was used to obtain the final quality score of stereoscopic images.The experimental results show that the algorithm framework boasts fairly sound accuracy and robustness.
作者
张爽爽
周武杰
ZHANG Shuangshuang;ZHOU Wujie(School of Information and Electronic Engineering,Zhejiang University of Science and Technology,Hangzhou 310023,Zhejiang,China)
出处
《浙江科技学院学报》
CAS
2020年第1期26-31,共6页
Journal of Zhejiang University of Science and Technology
基金
国家自然科学基金项目(61502429)
浙江省自然科学基金项目(LY18F020012)
关键词
图像质量评价
卷积神经网络
加权融合
融合图像
image quality evaluation
convolutional neural network
weighted fusion
fused image