Video Super Resolution Based on Motion Compensation

An image super-resolution method and device provided by acquiring the captured video of the power equipment. The method according to the time sequence of the image frames in the video of the power equipment, extracts the appearance image of the power equipment, selects the reference image, obtains the target image, and obtains the motion vector of any pixel moving from the position coordinates in the target image to the position coordinates in the reference image. The reference images are input into the spatiotemporal sub-pixel convolution network, and then the two-dimensional super-resolution image is output, which improves the image clarity, and can image the moving power equipment with high resolution. In this paper We use time fusion strategy in the model of video SR. Our model constructs a motion compensation mechanism based on spatial transformation network. This mechanism can be trained jointly for video SR. The results obtained by our method are proved to be superior in peak signal to noise ratio (PSNR).