Deformable alignment using random projections of landmark images

This paper presents a method for rigid alignment of objects undergoing deformation. Automated algorithms can be affected by auxiliary motion, such as image motion caused by transducer movement in echocardiography. Unlike de-formable registration methods, the goal of this work is alignment without introducing additional distortion. Our method, based on random projection theory, incorporates motion metadata for phase-aware alignment and outperforms rigid alignment approaches on synthetic data. We demonstrate the benefit of this as a pre-processing step to two common biomedical image analysis tasks: object segmentation and video denoising.