Reconstruction of transparent objects using phase shifting profilometry based on diffusion models

Reconstructing the 3D surface of transparent objects with phase shifting profilometry is difficult because refraction and internal reflections corrupt the sinusoidal fringe patterns at and near the object surface. Existing phase unwrapping and depth recovery algorithms assume clean fringe data and produce large errors or missing regions when the input is corrupted by transparency. This paper proposes using denoising diffusion probabilistic models (DDPMs) as a learned prior over plausible phase maps, conditioned on the observed corrupted fringe images. The iterative denoising process of the diffusion model naturally handles the inpainting and denoising sub-problems jointly, guided by the corrupted observation at each step. The method is evaluated on a diverse set of transparent objects including glass components and plastic parts common in industrial inspection. Results published in Optics Express (2024) show substantial improvement in reconstruction completeness and surface accuracy compared to classical phase retrieval and prior CNN-based approaches, establishing diffusion models as a promising tool for challenging optical measurement scenarios.

Problem setting

Transparent objects present a well-known challenge for phase shifting profilometry: refractions and specular reflections corrupt the captured fringe patterns in ways that violate the assumptions of standard phase retrieval algorithms, leading to missing or inaccurate depth regions. This work proposes using diffusion models—a class of generative models with state-of-the-art image synthesis capability—to reconstruct transparent object surfaces from corrupted profilometry data. The diffusion model is conditioned on the corrupted fringe observations and learns to iteratively denoise and inpaint phase maps to produce physically plausible surface reconstructions.

In the broader publication record, this work appears in Optics Express, 32(8):13342–13356. The visual notes below pair the paper’s original figures with a concise reading of the method, experimental setup, and reported results.

Method and visual evidence

The method follows an optical 3D measurement pipeline: acquire coded images, recover phase or geometric cues, compensate the dominant error source, and reconstruct a reliable 3D result.

The extracted figures below show the sensing setup, algorithmic signal flow, and representative reconstruction or calibration results.

Method overview. This image is extracted from an embedded PDF image object on page 7, then recomposed for web display.

Representation and setup. This image is extracted from an embedded PDF image object on page 8, then recomposed for web display.

Experimental evidence. This image is extracted from an embedded PDF image object on page 8, then recomposed for web display.

Result comparison. This image is extracted from an embedded PDF image object on page 11, then recomposed for web display.

Additional visual result. This image is extracted from an embedded PDF image object on page 12, then recomposed for web display.

Results and impact

The evaluation reported in Optics Express, 32(8):13342–13356 uses the extracted figures above to show the method’s measurement, reconstruction, segmentation, matching, or diagnostic behavior on representative experiments. These visuals are paired with the paper’s quantitative or qualitative analysis to make the workflow easier to inspect from the homepage.

Source handling

I extracted 46 candidate image objects from paper.pdf and generated the compressed WebP figures used on this page. The local PDF was also optimized from 4,100,222 bytes to 3,720,169 bytes.