A Depth Map Acquired By Integrated Imaging Based On Adaptive Matching Window

Xia Wen, Yu Wang, Yan PiaoAdvances 2015 · DOI

Abstract

This paper presents a novel method for obtaining high-accuracy depth maps using integrated imaging technology. The core innovation is an adaptive matching window that utilizes boundary information in the $I_1I_2I_3$ color space and a minimum threshold value to dynamically adjust its size and shape. To further enhance accuracy and reduce false matching rates, a three-view consistency checking method is integrated. Experimental results demonstrate that this approach significantly improves depth map precision, particularly at object boundaries.

Introduction: The Challenge of Stereo Matching

Stereo matching is fundamental to 3D imaging, involving the comparison of multiple images to determine parallax and, consequently, depth. A central challenge in this process is selecting the optimal "matching window"—the pixel block used for comparison.

Standard fixed-size windows face a "Goldilocks" dilemma. Large windows often overlap object boundaries, leading to "hypertrophy" where depth edges appear blurred or distorted. Conversely, small windows may lack sufficient texture information, resulting in noise and matching errors. While adaptive windows exist, they often rely solely on grayscale data, missing critical color-based edge information.

Proposed Methodology

The researchers propose a three-stage approach to overcome these limitations:

1. High-Resolution View Acquisition

Integrated imaging typically uses a microlens array to capture a scene from multiple perspectives. Instead of the traditional method of extracting a single pixel per lens, which yields low-resolution views, this method extracts $N \times N$ pixel blocks. This preserves more detail for the matching phase.

Integrated imaging setup using a microlens array to capture multiple perspectives.

2. Adaptive $I_1I_2I_3$ Color Space Window

To improve edge detection, the system transforms RGB data into the $I_1I_2I_3$ color space. This space is less correlated than RGB and more closely aligns with human visual perception. By applying a Canny operator within this space, the algorithm identifies precise boundaries. The matching window then "grows" from a central pixel, expanding until it hits these boundaries. This results in asymmetrical, geometry-aware windows that never cross depth planes.

3. Three-View Consistency Check

To mitigate errors from occlusions or repetitive textures, the system employs a three-view check. It compares a central reference view against both left and right views. If the calculated parallax values do not agree, the point is flagged as a mismatch and corrected using neighboring data with similar color profiles.

Three distinct views extracted from the lens array, used for the consistency checking algorithm.

Experimental Results and Discussion

The method was tested using a physical setup featuring a lens array ($f=3mm, p=1.08mm$) capturing a scene with printed text and logos (specifically a "Fly Eagle" brand package).

Key findings include:

  • Edge Preservation: The adaptive window effectively eliminated the "hypertrophy" effect, maintaining sharp, clean lines at object boundaries.
  • Error Reduction: The three-view consistency check successfully identified and corrected mismatched points, resulting in a smoother, more reliable depth map.
  • Color Sensitivity: Utilizing the $I_1I_2I_3$ color space allowed the system to distinguish boundaries that were nearly invisible in standard grayscale.

Conclusion and Future Work

The proposed adaptive window and consistency check provide a robust framework for high-quality depth map generation in integrated imaging. While the results are qualitatively superior, future work should focus on quantitative benchmarking against other state-of-the-art methods and optimizing the algorithm for real-time performance, perhaps through GPU acceleration.

Summary updated

Abstract

In this paper, we prose a method to obtain the depth map based on adaptive matching window by Integrated imaging technology. This method combines the boundary information in I1I2I3 color space and minimum threshold value of the window effectively. So that the self-adapted matching window could adjust its size and shape to improve the matching accuracy. Furthermore, in order to take advantage of multiple views of the integrated imaging, the consistency checking method is introduced into integrated imaging to reduce the false matching rate. Experimental results show that the proposed method can obtain the depth map with high accuracy.
BibTeX
@inproceedings{wen2015depth,
  author = {Xia Wen and Yu Wang and Yan Piao},
  title = {{A Depth Map Acquired By Integrated Imaging Based On Adaptive Matching Window}},
  booktitle = {Advances},
  year = {2015},
  doi = {10.2991/iccset-14.2015.85},
}
Figure