Image registration

Image Registration

For methods such as 4POLAR 2D or 4POLAR 3D, image registration is a mandatory step. This process aligns the four quadrants of your sensor—each capturing a different polarization angle ($0^\circ$, $45^\circ$, $90^\circ$, $135^\circ$)—so they perfectly overlay for analysis. PyPOLAR utilizes the Scale-Invariant Feature Transform (SIFT) algorithm to achieve sub-pixel alignment accuracy.

Registration Workflow

When the software detects a 4POLAR method, choose one of the following:

• Load Existing: If you have already registered this specific optical setup, click 'Load' to import a previously created registration file (*.pyreg).
• Perform New: If no registration exists, ensure you have a Beads image (*.tif) and a Whitelight image (must be named Whitelight.tif) in the same folder.
  • Click 'Perform' to run the registration.
  • Review the alignment figure. If accurate, click 'Validate'.
  • To reuse this alignment later, click 'Save' to generate a .pyreg file in the current folder.
  • If the result is unsatisfactory, click 'Change' to adjust the SIFT parameters.

---

Method Details

The registration pipeline follows three main stages:

1. FOV Segmentation: PyPOLAR analyzes Whitelight.tif using Otsu's thresholding and contour detection (cv2.findContours) to identify the boundaries of the four polarization fields.
2. Bead Extraction: The beads image is split into four separate images based on the Whitelight contours. Intensities are normalized and thresholded to isolate beads as distinct features.
3. Feature Matching:
   • The Top-Left quadrant is the "Fixed" (reference) image.
   • The other three quadrants are "Moving" images.
   • SIFT detects local features, and the software matches bead locations by solving a linear sum assignment problem (scipy.optimize.linear_sum_assignment).
   • A Homography Matrix (cv2.findHomography) is calculated to map the moving images onto the fixed image.

---

Parameters

Adjust these parameters if the software fails to find enough matching points or detects too much noise:

• contrastThreshold: Used to filter out weak features. Increase this value to ignore low-contrast noise; decrease it if valid beads are being ignored.
• sigma: The Gaussian blur applied at the initial stage. If your images are "soft" or captured with a low-resolution camera, reducing this number may improve feature detection.

See the OpenCV SIFT Reference for more technical details.

---

Troubleshooting: Creating a Whitelight File

If Whitelight.tif is missing, you can create one manually in Fiji/ImageJ to define the quadrant locations:

1. Open your Beads image in Fiji.
2. Adjust Brightness/Contrast (Image > Adjust > Brightness/Contrast) until the four quadrants are visible.
3. Use the Oval selection tool (hold Shift for a perfect circle) to draw a circle over the upper-left quadrant.
4. Right-click the selection and choose "Create Mask".
5. Move that selection to the next quadrant on the original image and repeat "Create Mask" until all four are done.
6. Invert the resulting mask image (Image > Lookup Tables > Invert LUT) so the circles are white on a black background.
7. Save the file as Whitelight.tif.

Example of a Fiji-built Whitelight mask: