Smartphone Light-Field Display Parallax Testing

ISEF Category: Technology Enhances the Arts

Subcategory: Display Technology  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

Your phone can act like a tiny window into a 3D scene. That sounds simple, but one bad lens alignment can turn the view into a blur. If you can measure how well a light-field display preserves horizontal parallax, you are testing the core trick behind convincing depth. That makes this a strong science fair project with real engineering stakes.

What Is It?

A light-field display tries to show slightly different images to your left and right eyes, or to your eyes at different positions, so a flat screen feels like a 3D scene. Think of it like a set of tiny peepholes. Each one shows a slightly different view, and your brain fuses those views into depth.

In this project, the display uses a 7×7 grid of smartphone-camera lenslets over an OLED panel. Each lenslet acts like a tiny optical channel. A small CNN, which means a convolutional neural network, predicts the view changes from a single input image. The goal is to see how well the output matches the expected motion parallax, which is the way objects shift as your viewing position changes.

You would not just ask, “Does it look 3D?” You would measure how close the synthesized views stay to the real viewpoint shift. That gives you a project that blends optics, display design, and computer vision.

Why This Is a Good Topic

This is a good science fair topic because you can test one clear thing, view accuracy, with images and controlled viewpoints. You do not need a giant lab to study the idea, but you do need careful design and data analysis. The project connects to real display problems, like virtual reality, wearable screens, and compact 3D imaging. You can learn optics basics, CNN evaluation, calibration, and image-quality metrics from a project that feels modern and original.

Research Questions

• How does lenslet alignment affect horizontal-parallax error across the display?
• What is the effect of viewpoint angle on synthesized image similarity to a reference view?
• Does a CNN-based view synthesis model outperform simple interpolation for edge regions?
• To what extent does OLED brightness variation change the measured parallax accuracy?
• Which lenslet grid spacing gives the best tradeoff between view separation and image sharpness?
• How does calibration error in the stepper rail affect repeatability of parallax measurements?
• What is the effect of different scene depths on visual quality and disparity consistency?

Basic Materials

• Smartphone or small OLED display panel.
• 7×7 lenslet array made from smartphone-camera lenslets or similar micro-lens elements.
• Stepper rail or linear motion stage for viewpoint sweep.
• Digital camera or smartphone camera for capture.
• Tripod or rigid mount system.
• Printed calibration target with high-contrast markers.
• Ruler or calipers for alignment checks.
• Computer for image analysis.
• Tape, clamps, and mounting frames.
• Black-out material or enclosure to control ambient light.

Advanced Materials

• Optical bench or rigid rail system.
• Micro-positioning stage for fine alignment.
• Photodiode or colorimeter for display brightness checks.
• High-resolution camera with manual settings.
• Diffusion and polarization control materials.
• Custom mount for repeatable lenslet-to-panel spacing.
• Reference display or calibrated monitor for comparison images.
• Optional depth camera for ground-truth scene geometry.
• Field-programmable controller or motion driver for sweep automation.
• Crosshair calibration target and slanted-edge target for imaging analysis.

Software & Tools

• Python: Processes captured images, computes parallax error, and runs statistics.
• OpenCV: Detects calibration markers and measures view shifts across frames.
• ImageJ: Measures sharpness, alignment, and contrast in captured images.
• Plotly: Makes clear plots of viewpoint position versus parallax accuracy.
• Matplotlib: Creates publication-style graphs for your final report.

Experiment Steps

1. Define the one display variable you will test first, such as lenslet spacing, alignment, or CNN input settings.
2. Build a calibration plan that maps each viewpoint position to a known rail location and a reference image.
3. Choose image metrics that turn visual quality into numbers, such as disparity error, edge sharpness, or structural similarity.
4. Set up controls that separate optical problems from software problems, so you can tell whether the lenslets or the CNN cause the error.
5. Plan a repeatability check with the same scene and the same sweep path, so you can measure consistency, not just one-off results.
6. Design your final comparison so you can rank configurations and explain which design choice improves parallax most.

Common Pitfalls

• Misaligning the lenslet grid with the OLED panel, which causes view crosstalk and weakens the 3D effect.
• Changing ambient light between trials, which shifts brightness and makes image comparisons inconsistent.
• Using a viewpoint sweep that is not repeatable, which blurs the relation between rail position and parallax error.
• Comparing synthesized views to the wrong reference frame, which makes the model look better or worse than it really is.
• Focusing only on visual appeal and skipping quantitative metrics, which leaves you without evidence for your claim.

What Makes This Competitive

A stronger project does more than build a display that looks cool. You would earn more credit by testing a precise question, using calibrated viewpoint data, and comparing at least two design choices or algorithms. A careful error analysis, especially one that separates optical misalignment from model error, can make the work much stronger. If you add a novel comparison across scene depths, lenslet geometries, or synthesis methods, your project starts to look like real research.

Project Variations

• Test the same display with synthetic scenes versus real photographs to see whether texture or depth complexity changes parallax accuracy.
• Compare a CNN-based synthesis pipeline with a non-neural baseline, such as interpolation or warping, to see which holds up better at the edges.
• Swap the scene type, such as faces, text, or fine line art, and measure which content breaks the display first.

Learn More

• MIT OpenCourseWare, Computer Vision: Search for lectures on image formation, camera models, and view synthesis.
• Stanford CS231n: Search the public course materials for convolutional neural networks and image-based learning.
• OpenCV Documentation: Find tutorials on calibration, feature detection, and image comparison tools.
• NIH PubMed: Search review articles on light-field imaging, display calibration, and perceptual quality metrics.
• NASA Earth Observatory: Use image analysis examples to practice studying contrast, alignment, and visual artifacts in real images.