UCLA Researchers Report a Deep Learning-Enabled Diffractive Display Design That is Based on a Jointly-Trained Pair of an Electronic Encoder and a Diffractive Optical Decoder to Synthesize Super-Resolved Images

Virtual and augmented reality (AR/VR) systems have gained a lot of attention in the last ten years. The main aim being making user experiences better in fields like human-computer Interactions, visual media art, etc. Holographic displays are a promising alternative that allows precise control and manipulation of the optical wavefront, simplifying the optical setup between the SLM and the human eye. Holographic displays use spatial light modulators (SLMs) and coherent illumination with, for example, lasers. Due to the limitations of the present wavefront modulator technology, holographic displays, in general, have relatively low space-bandwidth products (SBP) directly determined by the SLM’s total number of individually addressable pixels.

In order to overcome the SBP limitations imposed by the wavefront modulator or the SLM, the researchers at UCLA, presented a deep learning-enabled diffractive super-resolution (SR) image display framework based on a pair of jointly-trained electronic encoder and all-optical decoder. This framework projects super-resolved images at the output while maintaining the size of the image field-of-view (FOV). By encoding the high-resolution images (to be projected or displayed) into compact, low-resolution representations with fewer pixels per image, this diffractive SR display also enables a significant reduction in the computational burden and data transmission/storage, where k > 1 defines the SR factor that is targeted during training of the diffractive SR image display. 

They described a deep learning-enabled diffractive display architecture that uses wavefront modulators with poor resolution and a pair of jointly trained electrical encoders and diffractive optical decoders to create and project super-resolved images. The high-resolution images of interest are quickly pre-processed by the digital encoder, which is made up of a trained convolutional neural network (CNN), to convert their spatial information into low-resolution (LR) modulation patterns that are then projected via a low SBP wavefront modulator. With the help of thin transmissive layers constructed using deep learning, the diffractive decoder processes this LR-encoded data to all-optically synthesize and project super-resolved images at its output FOV. 

The outcomes show a super-resolution factor of 4 for this diffractive picture display, demonstrating a 16-fold increase in SBP. They also use 3D-printed diffractive decoders operating at the THz spectrum to experimentally verify the effectiveness of this diffractive super-resolution display. It is possible to scale up this diffractive image decoder to work at visible wavelengths, which encourages the development of big FOV and high-resolution screens that are portable, low-power, and computationally effective.

Hence this super-resolution capability of diffractive decoders could be used to project images of red, blue or green wavelengths. The building blocks of future 3D display technologies, such as head-mounted devices, may be derived from this diffractive super-resolution image display architecture for displays with improved resolution. This technology would be very useful to make user experiences in various fields more immersive and engaging.

Check out the Paper and Reference Article. All Credit For This Research Goes To Researchers on This Project. Also, don’t forget to join our Reddit page and discord channel, where we share the latest AI research news, cool AI projects, and more.