Super Resolution Reconstruction Technology of Film and Television Images Based on Convolutional Neural Network (CNN) Algorithm

With the continuous development of high-definition display technology, the demand for super-resolution reconstruction technology of film and television images in the fields of film and television production, video streaming, video surveillance, etc. is increasing. This paper studies the super-resolution reconstruction technology of film and television images based on convolutional neural networks, aiming to solve the shortcomings of traditional methods in high-frequency detail recovery and computational efficiency. In this paper, we make a comparison between bilinear interpolation, Lanczos interpolation, traditional super-resolution and SRCNN. Based on SRCNN, an end-to-end learning approach is used to automatically recognize high resolution and low resolution images. The network is composed of feature extraction, characteristic mapping and reconstruction. By training massive high-resolution images, the mean square error between the reconstructed image and the actual high-resolution image is reduced. Experiments show that the CNN algorithm can restore high-frequency details well, and obtain a peak signal-to-noise ratio of 39.27 dB and an SSIM of 0.9717 on Set5. At the same time, the CNN method also shows significant advantages in computational efficiency and resource consumption, with shorter running time, lower FLOPs and memory usage, and less energy consumption, making it suitable for application in resource-constrained environments.