🔬 optics

A Lensless Polarization Camera

This paper proposes a compact, single-shot lensless polarization camera that utilizes a diffuser and a striped polarization mask combined with a specialized reconstruction algorithm to recover four linear polarization images, offering a lightweight and cost-effective alternative to traditional multiplexing-based systems.

Original authors: Noa Kraicer, Shay Elmalem, Erez Yosef, Hani Barhum, Raja Giryes

Published 2026-03-19

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Noa Kraicer, Shay Elmalem, Erez Yosef, Hani Barhum, Raja Giryes

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to take a picture of a secret message written in invisible ink. Regular cameras can only see the "intensity" of the light (how bright or dark things are), but they can't see the "secret ink" (the polarization of light). Polarization is a hidden property of light waves, like the direction they are vibrating. Seeing this hidden layer helps us do cool things like see through fog, identify fake materials, or spot stress in transparent plastics.

The problem? Traditional cameras that see this hidden layer are usually huge, heavy, and expensive. They often need complex lenses or special sensors that sacrifice image quality.

This paper introduces a tiny, cheap, lensless camera that can see this hidden polarization layer in a single snapshot. Here is how it works, using some everyday analogies:

1. The "Frosted Glass" Trick (The Diffuser)

Instead of using a fancy glass lens to focus light (like a standard camera), this camera uses a piece of random frosted glass (called a diffuser).

The Analogy: Imagine shining a flashlight through a piece of crumpled wax paper. The light doesn't form a clear picture; it just scatters into a messy, random pattern of dots.
The Magic: In a normal camera, this mess would be useless. But this camera uses a super-smart computer algorithm. It knows exactly how that specific piece of wax paper scrambles the light. So, when the computer sees the messy pattern, it can mathematically "unscramble" it to reconstruct the original image. It's like having a puzzle solver that can rebuild a picture just by looking at the scattered puzzle pieces.

2. The "Striped Sunglasses" (The Polarization Mask)

Now, how do we see the polarization? The camera puts a striped mask right on top of the sensor (the part that catches the light).

The Analogy: Imagine the sensor is a giant grid of eyes. We put a pair of sunglasses over them, but instead of one color, the sunglasses have alternating stripes: some let through light vibrating horizontally, some at 45 degrees, some vertically, and some at 135 degrees.
The Result: Every single pixel in the camera now sees a slightly different "version" of the scene, filtered by the stripe it sits under. Because the "frosted glass" (diffuser) already scrambled the image, the computer can figure out which pixel belongs to which stripe and separate the four different polarization views from just one single photo.

3. The "Recipe" (The Reconstruction Algorithm)

The camera takes a photo of this scrambled, striped mess. The computer then uses a specific "recipe" (mathematical model) to fix it.

The Recipe: The recipe needs to know two things perfectly:
1. How the frosted glass scrambles the light.
2. Exactly how the striped sunglasses filter the light.
The Problem: If the recipe is perfect, the result is amazing. But if the real sunglasses are slightly different from the recipe (maybe the stripes are a tiny bit wider, or there's a tiny gap between the sunglasses and the sensor's eyes), the picture gets blurry or has weird artifacts.

What Did They Find?

The researchers built a prototype and tested it.

The Good News: When they simulated the camera on a computer where the "recipe" matched the "ingredients" perfectly, the results were fantastic. They could clearly see stress patterns in plastic bags and separate light sources that looked identical to the naked eye.
The Bad News: When they used the real physical camera, the images weren't quite as sharp. Why? Because in the real world, the mask wasn't perfectly glued to the sensor. There was a tiny gap (caused by the glass covering the sensor), which made the light blur slightly before hitting the pixels. This tiny mismatch confused the computer's recipe.

The Big Takeaway

This paper proves that you don't need a giant, expensive lens to take polarization photos. You can do it with a piece of frosted plastic and a simple striped sticker.

However, the secret to making it work perfectly in the real world isn't just writing better math; it's building better hardware. If we can glue that striped mask directly onto the sensor chip (like printing it right on the silicon), we can eliminate the tiny gaps and get perfect, high-quality polarization images in a camera that is as small as a coin.

In short: They built a tiny, lensless camera that sees the "invisible" polarization of light by using a scrambled lens and a striped filter, proving that with better manufacturing, we could have super-compact cameras for medical endoscopes, drones, and robots that can "see" things we currently can't.

1. Problem Statement

Polarization imaging provides critical information about scene structure, material properties, and light-matter interactions that is invisible to standard intensity-based cameras. However, existing polarization cameras face significant limitations:

Division-of-Focal-Plane (DoFP) Sensors: While they allow snapshot capture, they sacrifice spatial resolution and require complex demosaicing algorithms to mitigate artifacts.
Sequential Filtering: These preserve spatial resolution but require multiple exposures, making them unsuitable for dynamic scenes.
Metasurfaces and Stacked Architectures: These offer compact designs but rely on intricate, high-precision fabrication processes that are expensive and difficult to scale.
System Constraints: Applications like compact microscopy and endoscopy require extremely small, lightweight systems where conventional lenses are often precluded.

There is a need for a compact, low-cost, single-shot polarization camera that does not rely on complex optics or high-precision micro-fabrication, while maintaining high reconstruction quality.

2. Methodology

The authors propose a lensless polarization camera that replaces traditional lenses with a computational imaging approach. The system consists of three main components:

A. Optical Design

Random Diffuser: A diffuser (similar to DiffuserCam) is placed in front of the sensor. It spatially multiplexes the incoming light, spreading the response of each scene point across many sensor pixels. This creates a Point Spread Function (PSF) with many narrow, high-contrast features, enabling image reconstruction from partially sampled data.
Striped Polarization Mask: A simple mask composed of linear polarizer stripes with alternating orientations ( $0^\circ, 45^\circ, 90^\circ, 135^\circ$ $0^{\circ}, 4 5^{\circ}, 9 0^{\circ}, 13 5^{\circ}$ ) is placed directly above the sensor.
- This mask performs polarization multiplexing in the image plane.
- Unlike DoFP sensors which use microscopic pixels, this mask uses wider stripes (approx. 256 pixels wide in the prototype), leveraging the diffuser's ability to reconstruct from partial sampling.
Single-Shot Acquisition: The system captures a single diffused and polarization-multiplexed image containing information for all four linear polarization angles simultaneously.

B. Reconstruction Algorithm

The authors employ a physics-based reconstruction algorithm that explicitly models the forward process:

Forward Model: The measurement $y$ is modeled as a linear superposition of the scene's polarization components convolved with the diffuser PSF and modulated by the polarization mask:
$y_c = \sum_{p=1}^{P} S_p \odot (x_{c,p} * k_c)$
Where $x$ is the scene intensity, $k$ is the diffuser PSF, $S$ represents the mask transmission, and $\odot$ is element-wise multiplication.
Optimization: The inverse problem is solved using the Alternating Direction Method of Multipliers (ADMM). The objective minimizes a regularized least-squares function:
$\hat{x} = \arg \min_x \frac{1}{2\sigma_e^2} \|y - Ax\|_2^2 + \lambda \cdot \text{TV}(x)$
The regularization term uses weighted anisotropic Total Variation (TV) to promote image priors and handle the ill-posed nature of the problem.

3. Key Contributions

Novel System Architecture: Introduction of a compact, lensless polarization camera using a random diffuser and a simple striped polarization mask, eliminating the need for lenses and complex micro-polarizer arrays.
Single-Shot Recovery: Demonstration of recovering four linear polarization sub-images ( $0^\circ, 45^\circ, 90^\circ, 135^\circ$ ) from a single snapshot.
Systematic Mismatch Analysis: A rigorous investigation into model mismatch, identifying that the primary limitation in lensless polarization imaging is not the algorithm but the physical deviation between the assumed forward model and the real prototype (specifically mask-sensor distance and fabrication imperfections).
Prototype Validation: Successful demonstration of the system in both simulation (matched models) and real-world experiments (lab prototype), including applications like strain analysis in transparent materials.

4. Experimental Results

The evaluation was conducted in three stages:

Matched Forward-Model Simulations: When the reconstruction model perfectly matched the data generation model (using either measured or idealized masks), the system achieved high-quality reconstruction with high PSNR and SSIM. This proved the optical and algorithmic feasibility of the approach.
Controlled Mismatch Analysis: The authors introduced perturbations (Gaussian blurring, noise, and interpolation) to simulate real-world deviations.
- Finding: Even modest deviations in the polarization mask model caused a pronounced drop in reconstruction quality.
- Dominant Factor: The analysis identified polarization-mask mismatch as the dominant limiting factor. Specifically, the finite distance between the mask and the sensor (due to cover glass) causes spatial mixing and cross-talk between polarization stripes, which the idealized model fails to capture.
Real-World Prototype Experiments:
- The prototype successfully recovered polarization-dependent features, such as separating light sources based on polarization angle and visualizing birefringence-induced strain in a plastic bag under back-illumination.
- However, the real-world results showed reduced contrast and increased artifacts compared to simulations.
- The artifacts in the real prototype qualitatively matched those produced by the controlled "blur" perturbations, confirming that mask-sensor separation is the primary source of error.

5. Significance and Future Directions

Compactness: The system offers a path toward ultra-compact, lightweight polarization imaging suitable for endoscopy, microscopy, and drone applications where volume and weight are critical.
Insight into Limitations: The paper clarifies that the performance gap between simulation and reality is not fundamental to the lensless approach but arises from physical integration constraints.
Path Forward: The authors conclude that future improvements rely on tighter integration of the polarization mask with the sensor (e.g., on-chip or co-fabricated masks) to minimize the mask-sensor distance and fabrication tolerances. This would reduce model mismatch and allow the system to approach the ideal performance demonstrated in simulations.
Algorithmic Robustness: The study suggests that while learning-based priors might help, they are highly sensitive to forward-model inaccuracies; therefore, accurate physical modeling remains crucial for lensless systems with significant hardware variability.

In summary, this work demonstrates that lensless polarization imaging is a viable and promising technology, provided that future hardware designs prioritize precise mask-sensor integration to mitigate model mismatch.