Mitigating Optical Diffraction in Under-Display Cameras

32 concepts27 papers3 proposals

To achieve a truly bezel-less smartphone display, hardware engineers must place the front-facing camera directly beneath the active display panel. However, this configuration creates a severe optical conflict: the very structure that emits light to the user must also transmit light to the camera sensor. As shown by Marcos V. Conde et al. and Xina Liu et al., this semi-transparent barrier introduces severe optical interference, resulting in images marred by intense flare, pervasive haze, and a dramatic loss of contrast.

To solve this problem, researchers have turned to computational photography, combining physical optics modeling with deep learning to mathematically "dissolve" the display panel covering the lens.

The Display Panel as a Diffraction Grating

The primary reason under-display camera (UDC) images look blurry and hazy lies in wave optics. An active display panel (such as an OLED screen) consists of a regular, periodic grid of subpixels interspersed with microscopic gaps for circuitry and wiring. When light from the environment passes through these gaps to reach the underlying sensor, the grid acts as a diffraction grating.

As David Ren et al. demonstrate, the specific geometry of this grid dictates the nature of the image degradation. High-density pixel layouts cause severe light attenuation and chromatic shifts, while transparent panels tend to introduce highly directional blurring. Because the gaps act as tiny apertures, light waves interfere with one another, transforming sharp points of light into complex, structured flare patterns and streaks (Chengxu Liu et al.).

Furthermore, Hrishikesh Panikkasseril Sethumadhavan et al. emphasize that because these artifacts are tightly coupled to the hardware's physical layout, restoration algorithms must be tailored to the specific diffraction and noise characteristics of the underlying display panel.

Deconstructing the Degradation: Flare, Haze, and Scattering

To effectively invert this degradation, researchers break down the optical interference into distinct physical components. Chengxu Liu et al. decouple the artifacts into two primary phenomena: flare (the structured, long-range streaks caused by diffraction) and haze (the global loss of contrast).

This haze is heavily driven by what Binbin Song et al. identify as the scattering effect of the semi-transparent display layers. When light hits the internal boundaries of the display assembly, it scatters randomly, creating a uniform veil of background light that obscures fine textures and details.

Standard denoising and deblurring techniques fail here because they assume localized, spatially invariant degradations. UDC artifacts are inherently non-local; a bright light source in one corner of the frame can cast a diffraction flare or scattering haze across the entire sensor.

Reverting Diffraction in the Right Domains

How do modern computational photography pipelines undo these complex wave-optic effects? The key is choosing the right domain for restoration.

The Raw Sensor Domain

Traditional image restoration networks operate on standard sRGB images. However, before an image becomes an sRGB file, it passes through an Image Signal Processor (ISP) that applies non-linear operations like gamma correction, tone mapping, and demosaicing. These non-linear steps warp the mathematical relationship between the original scene and the captured diffraction pattern.

To circumvent this, Nam Ik Cho introduces a framework that restores images directly in the raw sensor domain. By learning an unsupervised "degradation fingerprint" of specific panels and operating on linear raw data, the network can accurately invert the physical process of diffraction before the ISP scrambles the signal.

The Frequency Domain

Because diffraction patterns are periodic and structured, they are often easier to isolate in the frequency domain than in the spatial domain. In the spatial domain, a diffraction streak covers many pixels and obscures fine facial or background textures (Jingfan Tan et al.). In the frequency domain, however, periodic structures manifest as distinct, localized peaks.

Chengxu Liu et al. exploit this by designing a Frequency and Spatial Interactive Learning Network (FSI). By jointly processing the image in both domains, the network can efficiently identify and eliminate the periodic frequency components responsible for the streaks while preserving the underlying spatial textures.

Designing for Real-World Mobile Constraints

While complex physical modeling and heavy deep-learning architectures can restore UDC images in a laboratory setting, deploying these solutions to commercial smartphones requires balancing restoration quality with strict computational budgets. Synthetic datasets often fail to capture the messy reality of physical hardware, a gap highlighted by Kyusu Ahn et al., who developed real-world paired datasets to better train networks for actual diffraction and flare.

To bridge the gap to real-time mobile execution, researchers have developed highly optimized network blocks. For instance, Chongyi Li introduces a Dynamic Skip Connection Network to adaptively restore degraded regions without wasting computation on unaffected areas. Similarly, the authors of Perceptual Image Enhancement for Smartphone Real-Time Applications utilize lightweight inverted residual attention blocks to process high-resolution images within strict mobile power budgets.

By combining joint hardware-software insights—as advocated in the broader computational photography community by Kavita Bala et al.—modern UDC systems successfully leverage deep learning to deliver clear images from behind an active screen.

Key Foundational Work

Symmetric Photography: Exploiting Data-sparseness in Reflectance Fields (2006) — contributed 1 concept elements to this direction.
The Solar Optical Telescope for the Hinode Mission: An Overview (2008) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_under_display_optical_interference_47a4cabd,
  title = {under-display optical interference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/under-display-optical-interference/rdr_47a4cabd787447ba9e88781ff155e5cb},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_under_display_optical_interference_47a4cabd_v1,
  title = {under-display optical interference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/under-display-optical-interference/rdr_47a4cabd787447ba9e88781ff155e5cb/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake