Machine Vision Assignment: Sinogram Image Reconstruction

📘 Overview

This project demonstrates image reconstruction from a parallel-projection sinogram using Python. The process includes:

• Unfiltered Backprojection
• Ramp Filtered Backprojection
• Hamming-Windowed Ramp Filter
• Hann-Windowed Ramp Filter

Each RGB channel is reconstructed separately and combined into the final image.

Contributors:

• Luke Griffin
• Patrick Crotty
• Michael Cronin
• Aaron Smith
• Cullen Toal

---

🧠 Background

Reconstruction from sinograms is a core concept in Computed Tomography (CT). This project builds upon:

• Radon Transform and Filtered Backprojection
• Frequency-domain filtering using Ramp, Hamming, and Hann windows
• Evaluation of visual clarity and artifact suppression in reconstruction

---

🏗️ Architecture & Workflow

1. Input Sinogram

• Load the RGB sinogram
• Extract metadata (e.g. aspect ratio)

🖼️ Sinogram Image

---

2. Preprocessing

• Separate into R, G, B channels
• Transform to frequency domain using FFT

---

3. Filtering

• Apply:
  • Ramp filter
  • Hamming-windowed ramp filter
  • Hann-windowed ramp filter

🖼️ Hamming Window Plot

🖼️ Hann Window Plot

---

4. Inverse Transformation

• Convert filtered data back to the spatial domain using inverse FFT

---

5. Backprojection

• Reconstruct image by backprojecting each filtered projection at its corresponding angle
• Accumulate results into a laminogram

---

6. Post-Processing

• Crop based on original aspect ratio
• Normalize pixel values
• Combine RGB channels into final image

---

7. Output

• Save reconstructed images for all filter types
• Display side-by-side comparisons

🖼️ No Filter Reconstruction

🖼️ Ramp Filter Reconstruction

🖼️ Hamming Ramp Filter Reconstruction

🖼️ Hann Ramp Filter Reconstruction

---

📊 Results Summary

• No Filter: Blurry image with strong artifacts
• Ramp Filter: Sharper image but some high-frequency noise
• Hamming: Balanced sharpness and noise suppression
• Hann: Smoother result, slight loss of detail

---

✅ Conclusion

Filtering is essential in sinogram-based image reconstruction to suppress noise and improve clarity. The Hamming window provided the best compromise between detail and artifact reduction.

---

📎 References

• Zeng, G. L. (2014). Model Based Filtered Backprojection Algorithm: A tutorial.
• Arai, Y. (2021). Local cone beam CT: how did it all start?
• IAEA Human Health Campus. 3D Image Reconstruction.
• Wikipedia. Radon Transform
• Wikipedia. Backpropagation
• ScienceDirect. Hamming Window