Stained and infrared image registration as first step for cancer detection

摘要

This paper presents a comparison between different image registration approaches to register infrared (IR) and hematoxylin and eosin (H&E) stained images. This registration will be needed as the first step of a new process based on mid-IR images, that will help clinicians in cancer early diagnosis, providing them with more information about the tissue under examination. The aim of the registration step is to translate the segmented areas of interest in cancer diagnosis from the H&E images to the IR images, reducing the process of segmentation to the stained images. Being able to perform an automatic segmentation of these regions, the amount of data to be analysed will be drastically reduced increasing the speed of the process. The results of this paper show that the projective transformation approach is the best option to register these kinds of images using a reasonable number of keypoints.