UNSUPERVISED DEFORMABLE IMAGE REGISTRATION METHOD USING CYCLE-CONSISTENT NEURAL NETWORK AND APPARATUS THEREFOR

摘要

Disclosed are an unsupervised learning-based image matching method using a neural network with cycle coherence and an apparatus therefor. An image registration method according to an embodiment of the present invention includes receiving a first image and a second image to be matched; outputting the transform fields for the first image and the second image by using an unsupervised learning-based neural network having cycle consistency with respect to the transform field; and generating a matched image for the first image and the second image based on a spatial transformation function to which the output transformation field is applied, wherein outputting the transformation field includes moving the first image. When the image is an image and the second image is a fixed image, outputting a transformation field for the first image for matching the first image to the second image, and generating the second image using the spatial transformation function The registered image may be generated by applying a deformation field of the first image to one image.