改进BasicVSR的水下视频超分辨率重构  

Improved Underwater Video Super-resolution Reconstruction Based on BasicVSR

作　　者：赵艳玲 张婧 冯迎宾 ZHAO Yanling;ZHANG Jing;FENG Yingbin(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang,China;Department of Mechanical and Electrical Engineering,Tongliao Industrial Vocayional School,Tongliao,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,沈阳 [2]通辽市工业职业学校机电工程系,通辽内蒙古

出　　处：《光电技术应用》2023年第5期66-73,共8页Electro-Optic Technology Application

摘　　要：由于水下环境复杂多变,水流湍急、相机抖动、悬浮颗粒的遮挡以及光的吸收和传播等导致光学设备获取的水下视频出现运动模糊、颜色失真和对比度低等问题。针对这些问题提出一种改进BasicVSR的水下视频超分辨率算法以提高重构图像的细节信息,同时改善水下图像偏蓝和偏绿的现象。首先,利用卷积神经网络拟合水下图像退化模型中的参数,进而得到水下图像特征;其次,利用原始输入的低分辨率视频帧计算得到前后帧之间的光流值;最后,利用光流信息对特征图进行双向传播,最终得到重构之后的每一帧图像。实验结果表明,经文中算法处理后的水下图像更符合人眼视觉特征,并且图像质量评价指标上相比其他算法有显著提高,更好的满足水下视觉高级任务对水下视频质量的要求。Due to the complex and variable underwater environment,turbulent flow,camera shake,occlusion of suspended particles and light absorption and propagation will lead to the problems of motion blur,colour distor-tion and low contrast of the underwater video.To solve these problems,an improved underwater video super-resolu-tion algorithm based on BasicVSR is proposed to enhance the details in reconstructed images,and at the same time improve the phenomenon of blue and green underwater image.At first,a convolutional neural network is used to fit the parameters of the underwater image degradation model and obtain the underwater image features.And then,the optical flow values between consecutive frames are computed using the low-resolution video frames as input.At last,the optical flow information is used for bidirectional propagation on the feature maps to reconstruct each frame.Ex-perimental results show that the underwater images processed by the algorithm exhibit better conformity to human visual characteristics and significantly improve image quality evaluation metrics compared with other algorithms,which better meet the requirements of underwater video quality for underwater advanced visual tasks.

关 键 词：水下视频超分辨率重构 卷积神经网络 水下图像退化模型 双向特征传播 光学设备 

分 类 号：TP394.11[自动化与计算机技术—计算机应用技术]