Super-resolution reconstruction of depth image based on edge-selected deep residual network

摘要

In recent years, depth images have been widely used in human daily life and production, but the quality of the images obtained is often poor and the resolution is low. The existing improved methods often assume that there are common edges between the color image and the depth image, and then super-resolution the depth image under the guidance of the color image. When the part of depth image is smooth and the high-frequency details of the corresponding part of color image are rich, the color details that should not exist will be introduced into depth image. In this paper, an edge-selected deep residual network is proposed for depth image super-resolution. The scheme can effectively extract the common features of color and depth images, reduce the influence of irrelevant color details on super-resolution images, and ultimately improve the resolution of depth images. In the experiment of super-resolution, this method is effective and reliable.