This paper includes several new things aimed at matching microscopy images with small overlapping areas. However, we now use the same descriptor (FFF) for matching word images. The image below shows an image where two images of a kidney sample from a Mini-TEM microscopy are being matched. Even if the overlap is very small the algorithm proposed managed to make a stitching. The red crosses are non matching points and the green ones are correct matches.

smallOverlap2

The paper describes improvements for the whole image stitching pipeline:

  1. Detection of interest points.
    • The Harris Corner detector can be simplified when rotation invariance is not required.
    • Instead of taking the strongest points it is proposed to use a regular sampling strategi where the image is divided into smaller areas and take the strongest points in each of them.
  2. Extraction of feature descriptors for each interest point.
    • Use one or two elements of the DFT instead of M3. We now use this in our word spotter when comparing word images.
  3. Matching of descriptors in order to find tentative correspondences.
    • We propose to use Cascade matching, which means that first a very small descriptor based on the feature vectors, is used in the Nearest Neighbour search in order to tell whether they possibly could be matching, before using the complete feature vector for matching.
  4. Removing false correspondences, e.g. using RANSAC, and computing the transformation between the pair of images.
    • We show that Clustering can be a reliable, deterministic and especially faster alternative to RANSAC.
  • A Fast Fourier Based Feature Descriptor and a Cascade Nearest Neighbour Search with an Efficient Matching Pipeline for Mosaicing of Microscopy Images. A. Hast, V. A. Sablina, I-M. Sintorn, G. Kylberg. Pattern Recognition and Image Analysis: Advances in Mathematical Theory and Applications in No 2., Journal. 2018.
    @article{Has18a, author = {A. Hast and A. Sablina and I-M. Sintorn and G. Kylberg}, title = {A Fast Fourier Based Feature Descriptor and a Cascade Nearest Neighbour Search with an Efficient Matching Pipeline for Mosaicing of Microscopy Images}, journal = {Pattern Recognition and Image Analysis: Advances in Mathematical Theory and Applications in No 2}, year = {2018}}

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s