Enhanced integral imaging 3D depth map by using SRCNN

In this paper, we propose a method that can generate a three-dimensional (3D) depth map accurately by using an integral imaging technique through low-resolution elemental images. The conventional method may produce an accurate 3D depth map when it utilizes high-resolution images. However, if the elemental image has low resolution, it can cause the shifting pixel error when it reconstructs the 3D image by using volumetric computational reconstruction (VCR). Therefore, it may not provide a high-quality 3D depth map through the reconstructed image. To solve this problem, our proposed method utilizes a machine learning technique that can enhance the resolution of the elemental image. Our proposed method enlarged the low-resolution elemental image. Then, it generates the high-resolution image through the super-resolution convolution neural network (SRCNN). Therefore, it can reduce the shifting pixel error and it can generate various reconstructed images according to the depth. Finally, it can generate an accurate 3D depth map through the error-reduced 3D image. To prove our proposed depth map accuracy, we implement the simulation experiment and evaluate the image through the image quality metric.