Novel Deep Learning Technique for Editing and Generating High-Quality Holograms
Researchers overcome noise and low resolution of extracted light field from conventionally generated holograms using a novel deep-learning method
Modifying holograms requires extracting light field data of the 3D object portrayed by the hologram. However, current methods for extracting this data result introduce noise and low resolution, degrading the final image quality. To address this, researchers employed a deep learning technique, which effectively eliminated noise and improved the quality of light field data extracted from holograms. This innovative method allows the editing of light field data and consequently the generation of high-quality holograms.
Holography is a fascinating technology that allows the generation of three-dimensional (3D) images, called holograms, with applications in a wide variety of fields. However, there are several challenges in the way of achieving commercial-grade holograms due to the lack of appropriate optical devices, among which, the editing of holograms is a key issue. To edit a hologram, accurate information on the 3D object portrayed by the hologram is required. Despite much research, techniques for extracting this information are still not known.
One way to extract the 3D object information is to acquire light field data instead of an actual 3D model. Light field data consists of light rays emanating from a 3D object, represented as a set of views from multiple angles. This light field data can then be modified and recreated as a hologram using a computer. In this process, the quality of the extracted light field data is the most important factor that influences the quality of the final modified hologram. Light field data can be extracted by passing the angular spectrum of the light from a 3D object through a band-pass filter. However, this method has several issues, including low spatial resolution, blurring and speckle noise, which degrade image quality during extraction.
To address these issues, Dr. Dae-youl Park from the Electronics and Telecommunications Research Institute in Korea employed an innovative new method for processing extracted light-field data. “We employed a deep learning technique to improve the quality of the light field, which compensates for the inevitable degradation in image quality of extracted light field and enables the creation of an edited hologram with the same quality as the original image”, explains Dr. Park. Their study was published in the ETRI Journal on 5 July, 2023.
To remove speckle noise and blurring, and to improve spatial resolution, the researchers employed a generative adversarial network based deep learning model. This model is comprised of two networks: a generator and a discriminator. The function of the generator model is to generate an image similar to the real image, while the discriminator compares the real and generated images. During the training of this model, the extracted light field data from the DIV2K image set was used as input to the generator to create a speckle and noise-free image and the comparisons provided by the discriminator were used to improve the overall performance of the generator.
The trained generator was then tested by generating holograms of various objects with different depth characteristics. Testing revealed that the model was able to generate high-quality holograms regardless of the hologram generation method or the position of the bandpass filter.
Highlighting the importance of the study, Dr. Park remarks: “Holographic technology will bring about significant changes for us. It will transition the way we convey information from the current method of displaying information on 2D screens to communication in a 3D space. We believe that our method will play a key role in this transition.”
Overall, this study marks a significant step in the advancement of holographic technology!
Reference
Title of original paper: Improving the quality of light-field data extracted from a hologram using deep learning
Journal: ETRI Journal
About the Electronics and Telecommunications Research Institute (ETRI)
Established in 1976, ETRI is a non-profit government-funded research institute in Daedeok Science Town in Daejeon and is one of the leading research institutes in the wireless communications domain. It has filed more than 2500 patents. Equipped with state-of-the-art labs, this institute strives for social and economic development through technology research.
About the author
Dae-Youl Park received his PhD degree in electrical and computer engineering from Inha University, Incheon, Republic of Korea, in 2022. Since 2022, he has been a member of senior researcher at the Digital Holography Research Section at the Electronics and Telecommunications Research Institute (ETRI), Daejeon, Republic of Korea. His current research interests include computer-generated holograms, self-interference incoherent digital holography and artificial intelligence.
