Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a learned generalization of phase correlation that allows for the direct measurement of transformations between observations in a flexible manner, overcoming the limitations of fixed basis representations. By learning the basis for transformation decomposition, the method can handle both dense non-rigid deformations and unitary dynamics. The framework demonstrates superior performance on the ACDC cardiac-MRI benchmark and matches state-of-the-art results on CAMUS echocardiography, while also successfully recovering eigenstates and energy levels in quantum systems from observation pairs.
Learning the basis for transformation decomposition unlocks new capabilities in image registration and quantum state recovery, outperforming traditional methods.
Correspondence is fundamentally relational: it seeks the unknown transformation between two observations of a common scene, not the content of either. Yet the dominant learning-based methods do not represent the transformation as a first-class object in the architecture. They encode each image independently and let a learned similarity function or a deep decoder discover the mapping implicitly. Phase correlation is the canonical exception, measuring the inter-image relationship directly in the Fourier domain, but the rigidity of its fixed basis confines it to global translation. We introduce a learned generalization of phase correlation that lifts this restriction by learning the basis on which the transformation decomposes. The same algebraic primitive extends to dense non-rigid deformations and to unitary dynamics. On the ACDC cardiac-MRI benchmark the framework matches or exceeds prior published baselines on both registration directions. On CAMUS echocardiography it matches state-of-the-art without auxiliary scoring or adaptive-smoothness mechanisms. Applied to time-evolved wavefunction pairs of the 1-D quantum harmonic oscillator, the same framework recovers the Hermite-function eigenstates and the quantized energy levels of the unknown Hamiltonian from observation pairs alone.