基于卷积神经网络的超分辨率失真控制图像重构研究

目的 解决超分辨率图像重构模型中存在的功能单元之间关联性差，图像色度特征提取完整性不强、超分辨率重构失真控制和采样过程残差控制偏弱等问题。方法 通过在卷积神经网络模型引入双激活函数，提高模型中各功能单元之间的兼容连接性；引用密集连接卷积神经网络构建超分辨率失真控制单元，分别实现对4个色度分量进行卷积补偿运算；将残差插值函数应用于上采样单元中，使用深度反投影网络规则实现超分辨率色度特征插值运算。结果 设计的模型集联了内部多个卷积核，实现了超分辨率色度失真补偿，使用了统一的处理权值，确保了整个模型内部组成单元的有机融合。结论 相关实验结果验证了本文图像重构模型具有良好可靠性、稳定性和高效性。

Image Reconstruction of Super-resolution Distortion Control Based on Convolutional Neural Network

The work aims to solve problems of poor correlation between functional units, weak completeness of image chromaticity feature extraction, weak distortion control in super-resolution reconstruction, and residual control in sampling process in super-resolution image reconstruction models. By introducing the double activation function into the convolutional neural network model, the compatibility and connectivity between the functional units in the model were improved. A super-resolution distortion control unit was constructed using a dense connected convolutional neural network to perform convolutional compensation operations on four chromatic components, respectively. The residual interpolation function was applied to the upsampling unit and deep backprojection network rules were used to achieve super-resolution chromaticity feature interpolation operations. The designed model set combined multiple convolutional kernels internally to achieve super-resolution chromaticity distortion compensation. A unified processing weight was used to ensure the organic fusion of the internal components of the entire model. In conclusion, the relevant experimental results verify that the image reconstruction model proposed in this paper has good reliability, stability, and efficiency.