Large Scale Image Registration

Image registration is to align a pair of misaligned images together. Traditional image alignment algorithms involve keypoint extraction, matching, and transform estimation. However, they generally failed on large scale image as the matched keypoints are not distributed correlatively with the image texture. Based on that, we proposed a tile-based feature extraction method for large scale textured based image, which enforces a uniformed sampling of keypoint detection on all region of the images. We also used parallel programming to accelerate the speed of descriptors’ extraction and matching by 4-5 times compared to the traditional method.

Keywords: Image Registration, Tile, Parallel Programming, Multiprocessing, Keypoint Extraction

  • Collaborated with NINDS, Nation Institute of Health
  • Contributed to the whole brain image analysis probject published in Nature Communication
  • Advisors: Dr.Dragan Maric, Dr.Badri Roysam
  • Code: [Python]

  • Maric. D. Jahanipour.J, Li, X.R., et al. (2020). Comprehensive Cell Phenotyping Method for Whole-Brain Tissue Mapping Using Highly Multiplexed Immunofluorescence Imaging, Computational Reconstruction, and Deep Neural Networks, Nature Communications (Accepted)