• Creates a large composite image from a series of smaller images
• Useful when the specimen is larger than the microscope’s or camera’s field of view
  • Useful for metallurgical, plant and tissues samples
• Changing magnification to increase field of view may lose resolution and remove important details in the image
• Use a scanning stage to acquire image tiles
  • Synchronize image capture, stage movement and order of images captured

• Two steps to create a composite image
  • Compute the overlap between the image tiles
  • Stitch (combine) the image tiles
• Compute overlap either manually or automatically
  • Region based or point based
• Stitch images using different schemes
  • Overlap, weighted average etc.

Computing Overlap — Region Based

• Select a region in the left image.
• Find matching region in the right image
  • Simple Sum-of-Differences method is an effective and fast method for many applications
  • Use correlation or correlation based methods for challenging situations
  • For images dominated by strong edges, use the gradient (edge) information in the image for matching

Computing Overlap — Region Based Example

1. Move right image over left image starting an offset determined by overlap estimate
2. Compute matching metric over overlapping region
   • Sum-of-Difference (SoD) or Correlation
3. Shift right image by a pixel from top-left to bottom-right of search region
4. Find point with smallest SoD or largest correlation
   • Use interpolation to find subpixel location

Computing Overlap — Point Based

• Solve a set of linear equations to estimate overlap (affine transformation) parameters
  • P_{Left Image} = AP_{Right Image} + T
    • where P is a set of points
    • T represents the translation in x and y directions
    • A is the 2×2 affine transformation matrix that contains rotation and scale
  • Example: Translation + Rotation

Computing Overlap — Point Based Example

1. Select a point of interest
2. Find neighboring pixel values — for example a 3×3 or 5×5 neighborhood
3. Define search region in second image — use rough estimate of overlap to define the region
4. Perform a correlation in the search region to find matching point — correlation peak
5. Repeat process for other points of interest

Stitching Images

• Overlap second image over the first

Stitching Images

• Use a weighted average to compute combined image pixel values