Stabilize Image Processing Pipeline (DL + CV) for Tomography / Ophthalmic Imaging

📘 Description

We need an experienced Computer Vision / Machine Learning Engineer to improve and stabilize our image processing pipeline so that it works reliably on images of varying quality and from different smartphones.

📌 Goal

Ensure the algorithm never crashes on poor-quality frames, handles variability confidently, and returns clear statuses (OK / LOW_QUALITY / FAIL / RETAKE) for downstream analysis.

✅ What Needs to Be Done

1. Unified Image Preprocessing

Create a single preprocessing function to be used in all modules (both Deep Learning and classical CV):

• Convert to a unified format (uint8, consistent color)
• Unified resizing
• Contrast normalization (CLAHE or percentiles)

• Deliverable: preprocess_input(img)
  
  2. Deep Learning as Main, Classical CV as Fallback
  
  Implement a robust strategy:
  
  

  • First try DL-based segmentation (center + rings)
  • If DL confidence is low — fallback to classical CV
  • If both fail — return status “RETAKE”

• Deliverable: A stable decision workflow integrating DL and CV

  
  

  3. Remove Hardcoded Parameters

  Replace fixed radii, thresholds, and block sizes with parameters that depend on image size and metadata.

  Deliverable: Parameter generator based on image shape

4. Add Quality Checks & Crash Protection

Ensure:

• No crashes when masks are empty or circles are missing
• Function returns clear statuses: OK / LOW_QUALITY / FAIL / RETAKE
• Graceful handling of unexpected inputs

5. Quality Gate Before KISA Calculation

Implement a gate that:

• Only computes KISA when center and quality are reliable
• Otherwise skips KISA and requests a new image

Deliverable: Quality gating module

📦 Expected Outputs

✅ One stable, unified pipeline for preprocessing + detection + status output

✅ Clear status codes for production use

✅ Parameterization for robustness across devices

✅ Well-documented code and minimal test

🎯 Skills Required

You should have:

✅ Strong CV + DL experience (PyTorch / TF)

✅ Experience combining classical CV with neural networks

✅ Experience writing production-grade, crash-safe pipelines

✅ Ability to parameterize and generalize methods

✅ Knowledge of medical imaging and quality metrics (preferable

📍 What We Provide

📌 Access to labeled images

📌 Current prototype pipeline

📌 Evaluation criteria and quality gates

📌 Remote collaboration