Thumbnails:
List:
Year:
Category:
Session:
Poster:
Getting poster data...
Neha Goyal, Yahiya Hussain, Gianna Yang, Daniel Haehn (University of Massachusetts-Boston, Boston, US)
In Connectomics, researchers are creating the wiring diagram of the brain at nanoscale. For this, 2D EM images need to be aligned to 3D volumes. To investigate if adding biological features can improve existing alignment methods, we use mitochondria masks data to guide the registration procedure in real-time. Automatic registration was performed on randomly rotated unaligned EM images with the help of conventional feature matching methods. We propose a new feature detection method, MITO, that uses biological features to create the bounding boxes and relocate those on image data. For each bounding box, we calculated an approximate polynomial that contains a list of (x, y) coordinates which are now considered as keypoints. We found that, with additional biological features, the overall execution time decreased by 27%. With MITO, the aligned images are obtained in real time with a throughput greater than 33 MP/sec and with dice score greater than 0.89 for the entire image set in less than 12 seconds. We present detailed analysis for the following feature generators ORB, BRISK, FAST, FREAK, MITO(ours).https://github.com/nehagoyal1994/mito/tree/main/image_registration