Multimodality registration for ocular multispectral images via co-embedding

Image registration of sequential multispectral images plays a vital role in retinal image analysis, since the appearance of ocular tissues significantly relates to the diagnosis, treatment, and evaluation of various diseases in ophthalmology. State-of-the-art multimodality image registration techniques greatly rely on mutual information between paired images to obtain their correspondence. However, it has been observed that mutual information-based image registration approaches suffer from inaccuracy especially when they are applied to small-sized images. Bearing this in mind, a novel groupwise registration approach is proposed by mapping the extracted features from multimodality images into the same latent space. To evaluate the proposed approach, the comparison experiments are conducted between state-of-the-art methods and the proposed approach. Experimental results demonstrate the superior accuracy of the proposed approach over the state-of-the-art techniques. Therefore, the proposed algorithm could be an invaluable tool for multimodality image registration applications.