摘要
随着机器学习在图像识别领域应用的深入,近年来衍生出了大批如NSS-TS/BRISQUE等图像质量评价方法,但其在计算精度和性能方面相对较差。因此,作者提出了基于卷积神经网络的图像质量评价算法,通过对图像的预处理、层次化分池处理等方法,有效提升了算法的学习能力和评价精度。通过与NSS-TS/BRISQUE/BLIINDS-II等算法的对比,证明算法在线性相关性和Spearman等方面优势明显。
With the application ofmachine learning in the field of image recognition, a large number of image quality evaluation methods,such as nss-ts / BRISQUE, have been derived in recent years, but their calculation accuracy and performance are relatively poor. Therefore,this paper proposes an image quality evaluation algorithm based on convolutional neural network. Through image preprocessing,hierarchical pool processing and other methods, it effectively improves the learning ability and evaluation accuracy of the algorithm.Compared with nss-ts / BRISQUE/ bliinds-ii and other algorithms, the algorithmhas obvious advantages in linear correlation and Spearman.
作者
李慧
LI Hui(Department of Information Technology,Anhui Food Engineering Vocational College,Hefei 230001,China)
出处
《遵义师范学院学报》
2021年第5期84-87,共4页
Journal of Zunyi Normal University
关键词
卷积神经网络
图像质量评价
线性相关
convolutional neural network
image quality evaluation
linear correlation
