【摘要】 图像的超分辨率重建过程仅仅对单幅图像进行处理,而视频通常被理解为基于时间轴排列的序列图像,这种时空相关性导致相邻帧间的图像往往包含大量的互补信息。在进行视频超分辨率重建算法中,通过结合当前帧与相邻帧来共同进行卷积神经网络的训练,可以充分获取相邻帧间的互补信息,达到提高超分辨效果的作用。同时利用超分辨率后的图像与真实图像间的运动轨迹具有一致性,来作为算法的损失函数。更多还原

【Abstract】 The super-resolution reconstruction process of the image only processes a single image,but the video is usually understood as a sequence image arranged based on the time axis. This space-time relativity causes the images between adjacent frames to often contain a large amount of complementary information. In the video super-resolution reconstruction algorithm,the convolutional neural network training is performed jointly by combining the current frame and the adjacent frames,and the complementary information between adjacent frames can be fully obtained,thereby improving the effect of super-resolution. At the same time,the consistency is used between the super-resolution image and the real image as the loss function of the algorithm.更多还原