Multi-View Image Reconstruction Algorithm Based on Virtual Reality Technology

Abstract

In this paper, dense convolutional self-coding block is proposed to extract multi-scale spatial features, and three connection modes are designed to realize feature reuse: intra-unit jump connection and inter-unit dense connection to realize short-term feature reuse mechanism; The inter-block hop connection realizes the long-term feature reuse mechanism. In this paper, a deep convolutional neural network for image super-resolution reconstruction is constructed based on dense convolutional self-coding block, and the network parameters are reduced through multi-scale supervised training, further improving the image reconstruction quality.