A novel approach has been proposed by researchers from Chiba University, using neural networks to transform 2D colour images into 3D 'holograms'.
The approach could simplify 3D 'hologram' generation and find applications in numerous fields, including healthcare and entertainment.
The proposed scheme comprises three deep neural networks and could infer 3D 'holograms' directly from a 2D image without requiring depth information, with the final 'hologram' containing a natural depth cue.
'Holograms' are traditionally constructed by recording the three-dimensional data of an object and interpreting the interactions of light with the object. This is considered to be computationally intensive, requiring the use of a special camera to capture 3D images.
Other deep learning methods have been proposed for generating ‘holograms’, offering the potential to create ‘holograms’ directly from the 3D data captured using RGB-D cameras that capture both colour and depth information of an object. This approach avoids computational challenges associated with the conventional method, offering a potentially easier way to generate ‘holograms’.
The new approach, proposed by Professor Tomoyoshi Shimobaba, could see 3D images created directly from regular 2D colour images captured using ordinary cameras, using three deep neural networks to make use of colour images as the input. The networks could then predict the associated depth map, providing information about the 3D structure of the image, using one network to map depth, a second to generate a ‘hologram’ and third DNN to refine the hologram generated by the second network. This approach could make the generated ‘holograms’ suitable for display on different devices.
Shimobaba commented: “There are several problems in realizing holographic displays, including the acquisition of 3D data, the computational cost of holograms, and the transformation of hologram images to match the characteristics of a holographic display device. We undertook this study because we believe that deep learning has developed rapidly in recent years and has the potential to solve these problems.
“Another noteworthy benefit of our approach is that the reproduced image of the final hologram can represent a natural 3D reproduced image. Moreover, since depth information is not used during hologram generation, this approach is inexpensive and does not require 3D imaging devices such as RGB-D cameras after training.”