摘要
图像质量的优劣直接影响数据分析和决策的准确性。因此,为提高数字图像质量,解决图像中的模糊和噪声问题,研究提出一种基于多尺度残差网络的轻量级双输出网络结构。实验数据显示,在GoPro数据集的测试中,该方法达到了33.42 dB的峰值信噪比和0.962的结构相似度,同时,其处理速度为0.774 s,模型参数量控制在1.097 M。在RealBlu r数据集上,该方法同样展现出了超越现有技术的性能。研究提出的轻量级双输出多尺度残差网络为数字图像质量提升领域提供了一种有效的解决方案,其在理论和实践层面的贡献都具有重要的实际意义和应用价值。
Image quality directly affects the accuracy of data analysis and decision making.Therefore,in order to improve the quality of digital images and solve the problems of blur and noise in images,a lightweight dual-output network structure based on multi-scale residual network is proposed.Experimental data show that in the test of GoPro dataset,the proposed method achieves a peak signal-to-noise ratio of 33.42dB and a structural similarity of 0.962.At the same time,the processing speed is 0.774s,and the number of model parameters is controlled at 1.097M.On the RealBlur dataset,the method also shows performance beyond existing techniques.The lightweight dual-output multi-scale residual network proposed in this paper provides an effective solution in the field of digital image quality improvement,and its contribution in both theory and practice has important practical significance and application value.
作者
谷诚
周学军
GU Cheng;ZHOU Xuejun(School of Physics and Electronic Information,Yan'an University,Yan'an Shaanxi 310700,China)
出处
《佳木斯大学学报(自然科学版)》
2025年第4期15-18,共4页
Journal of Jiamusi University:Natural Science Edition
关键词
数字图像
多尺度残差
质量提高
特征融合
digital image
multi-scale residuals
quality improvement
feature fusion
