Perceptual photometric seamlessness in projection-based tiled displays

Aditi Majumder, Rick StevensACM Trans. Graphics 2005 · DOI

TL;DR

This paper tackles the challenge of creating seamless, high-quality images on large tiled displays made of multiple projectors. Instead of aiming for perfect photometric uniformity (which degrades image quality), the authors propose a perceptually driven approach that exploits human visual limitations to achieve seamlessness with better dynamic range.

What Problem Does This Paper Address?

Large-scale scientific visualizations and virtual reality applications often rely on tiled displays composed of multiple projectors to achieve high resolutions. A major hurdle in creating these displays is achieving "photometric seamlessness," meaning the brightness and color appear uniform across the entire display, without visible seams between individual projectors. Previous approaches often focused on strict photometric uniformity, demanding that every pixel on the display have identical brightness response. However, this strict requirement often leads to severely compressed dynamic range and poor overall image quality. The authors argue that this strict uniformity is unnecessary and that a perceptually uniform display, which accounts for human visual perception, is sufficient and allows for higher image quality.

Methods

The paper introduces a three-step algorithm:

  1. Reconstruction: This offline step involves estimating the photometric characteristics of each projector. This includes:
  2. Transfer Functions ($h_l$): These describe how input intensity maps to output luminance for each color channel (red, green, blue). They are estimated using a photometer.
  3. Luminance Functions ($W_l, B$): These capture the spatial variation in maximum luminance ($W_l$) and black offset ($B$) across the projector's surface. These are estimated using a digital camera and specialized test images.
  4. Modification: This is the core of the paper's contribution. Instead of aiming for strict uniformity, the authors define a "perceptual uniformity" goal. This involves:
  5. Common Transfer Function ($H_l$): A single, spatially invariant transfer function is chosen for all projectors, replacing individual projector transfer functions.
  6. Display Luminance Functions ($W'_l, B'$): The reconstructed projector luminance functions are combined and then "smoothed" using a constrained gradient-based optimization method. This smoothing aims to minimize perceptible luminance variations, exploiting human visual limitations (e.g., contrast sensitivity function). The goal is to make these functions perceptually uniform, not necessarily geometrically flat.
  7. Display Quality Objective: The optimization also aims to maximize the display's dynamic range.
  8. Reprojection: This is the online, per-projector correction step. The input image for each projector is modified to compensate for the differences between the original projector characteristics and the desired hypothetical seamless display characteristics. This is achieved by applying the common transfer function, then the derived projector smoothing maps (scaling and offset), and finally the inverse of the projector's transfer function. This entire process is implemented using commodity graphics hardware (pixel shaders) for real-time performance.

Datasets

Not discussed in the paper. The authors use their own experimental setups at Argonne National Laboratory for demonstrations.

Key Insights and Ideas

  • Perceptual Uniformity vs. Strict Photometric Uniformity: The most significant insight is that strict photometric uniformity is overkill for achieving seamlessness. Human vision can tolerate smooth variations, and focusing on perceptual uniformity allows for better image quality and dynamic range.
  • Exploiting Human Vision: The contrast sensitivity function (CSF) of the human eye is directly used to define the smoothing criteria, ensuring that luminance variations are smoothed to a degree imperceptible to humans.
  • Optimization for Seamlessness and Quality: The problem is framed as an optimization problem balancing perceptual uniformity with maximizing dynamic range, rather than just achieving flatness.
  • Unified Approach: The method addresses intra-projector, inter-projector, and overlap region variations in a single, unified framework, unlike many prior edge-blending techniques.
  • Real-time Implementation: The ability to perform these corrections in real-time using commodity graphics hardware is a crucial practical advancement.

Key Results

  • The method was successfully demonstrated on tiled displays of various sizes: 2x2 (4 projectors), 2x3 (6 projectors), and 3x5 (15 projectors).
  • Qualitative results show significantly improved seamlessness and visual quality compared to uncorrected displays and displays corrected using strict photometric uniformity. Figure 7 visually demonstrates the trade-off: strict uniformity (bottom image) results in a washed-out appearance and reduced dynamic range, while the proposed method (second and third images) achieves seamlessness with better quality.
  • The real-time implementation achieves per-pixel correction with a frame rate reduction of only 10-20 frames per second, making it practical for interactive applications.
  • The algorithm is described as efficient, automatic, and scalable.

Limitations and Future Work

  • Chrominance: The current work primarily focuses on photometric (luminance) seamlessness. While the authors suggest that similar luminance function shapes across channels can mitigate chrominance seams, they acknowledge that explicit handling of chrominance variations is a future direction.
  • Non-Lambertian Displays: The model assumes Lambertian displays. While the correction is accurate from the camera's viewpoint, the seamlessness might degrade for viewing angles significantly different from the calibration viewpoint, though it performs well for a range of angles.
  • Gamut Mapping: The method does not explicitly address gamut mapping differences between projectors.
  • Self-Calibrating Systems: The authors suggest the need for self-calibrating systems that can correct themselves in real-time from arbitrary projected images, particularly for defense applications.
  • Perceptual Metric: A need for a quantitative perceptual metric to evaluate color correction methods is identified.
  • Smoothing Parameter ($\lambda$): The choice of the smoothing parameter $\lambda$ is crucial and depends on viewing distance and display resolution. For small images in print, a very large $\lambda$ would be needed to simulate infinite distance, which is not practical.

Summary updated

BibTeX
@article{majumder2005perceptual,
  author = {Aditi Majumder and Rick Stevens},
  title = {{Perceptual photometric seamlessness in projection-based tiled displays}},
  journal = {ACM Trans. Graphics},
  year = {2005},
  doi = {10.1145/1037957.1037964},
}
Figure