Unsupervised Multimodal Image Registration by Polynomial Warping over Correlation-Maximizing Shifts

Abstract

Alignment of images from multiple modalities is a very important procedure for many medical and industrial applications. Often times it is not possible to utilize a supervised method due to the lack of labeled data for any specific sensor architecture. In this study, a new unsupervised approach is proposed for a sensor-camera system aligned in one axis, that warps the image-like frames onto each other with a second-degree polynomial sampled from the cross-correlation maximizing segment shifts. This methodology will allow the registration process to adjust for focal differences and varying image modalities between the sensors. Thus, novel architectures utilizing seldom-used sensors will more easily adapt to industrial and medical work environments.