摘要
可见光图像重构质量评价一直是一个难点,因此,设计了基于深度学习网络的可见光图像重构质量评价方法。通过卷积神经网络(CNN)与图像质量评价方法(IQA)相结合,构成IQA-CNN模型,引入信息熵构建改进IQA-CNN模型,向该模型内输入重构可见光图像,归一化预处理后划分成数个分块,经有监督学习法训练该模型后,获取到该模型的最优参数,给定一幅重构可见光测试图像,运用训练后的改进IQA-CNN模型,获得该图像的质量评价分数,实现重构可见光图像的质量评价,结果表明,该算法的最优卷积核数量与采样步长分别为40个和90,在此最优参数下的评价性能指标SROCC与PLCC平均值:分别为0.977 5与0.974 2,所得评价结果与主观观测结果相吻合,可靠性与合理性更高。
The quality evaluation of visible light image reconstruction is always a difficult problem. Therefore, a visible light image reconstruction quality evaluation method based on deep learning network is designed. By combining convolutional neural network(CNN) with image quality assessment(IQA), the iqa-cnn model is constructed. The improved iqa-cnn model is constructed by introducing information entropy. The reconstructed visible light image is input into the model, which is divided into several blocks after normalization preprocessing. After the model is trained by supervised learning method, the optimal parameters of the model are obtained, Given a reconstructed visible light test image, the improved iqa-cnn model after training is used to obtain the quality evaluation score of the image and realize the quality evaluation of the reconstructed visible light image. The results show that the optimal convolution kernel number and sampling step size of the algorithm are 40 and 90 respectively, and the average values of srocc and PLCC under the optimal parameters are 0.977 5 and 0.974 2 respectively, The evaluation results are consistent with the subjective observation results, and have higher reliability and rationality.
作者
杨国梁
苏俊宏
薛鹏翔
李媛
YANG Guoliang;SU Junhong;XUE Pengxiang;LI Yuan(School of Optoelectrics Engineerring,Xi'an Technological University,Xi'an 710021,China)
出处
《激光杂志》
CAS
北大核心
2022年第1期95-100,共6页
Laser Journal
基金
陕西省科技计划项目-重点研发计划(No.K20180076)。
关键词
深度学习网络
可见光图像
重构质量
卷积神经网络
图像质量评价
信息熵
deep learning network
visible image
reconstruction quality
convolutional neural network
image quality evaluation
the information entropy
