Depth Image Resolution Enhancement using Discrete Wavelet Transform and Convolution Neural Networks

The depth image plays an increasingly important role in fundamental research and daily applications, with the reducing the price and increasing the number of affordable and portable depth cameras. Infrared sensors or depth sensors are widely used to control dynamic and static 3D scenes. However, the depth image quality is limited to low-quality images, as the infrared sensor does not have high resolution. Therefore, given the problems and the importance of using 3-D images, the quality of these images should be improved in order to provide accurate images from depth cameras. In this paper a resolution enhancement method of depth images using convolutional neural networks is considered. A convolutional neural network with a depth of 20 and three layers and a pre-trained neural network is used. We developed the system and tested its performance for two datasets, Middlebury and EURECOM Kinect Face. Results show for EURECOM Kinect Face images, PSNR improvement is approximately 7 to 16 dB and for Middlebury images the PSNR improvement is about 6 to 12 dB.