Medical image ensemble registration based on Gaussian mixture model and color component regularization

Ensemble registration has proven to be more robust and accurate than competing pairwise registration methods. Color medical images are commonly applied in medical diagnosis and analysis. In this paper ensemble registration using a Gaussian mixture model is extended from gray images to color medical images. To decrease the transformation differences among color component deformations of the same image, the color component regularization term is incorporated into ensemble registration and a novel total cost function is proposed. Color ensemble registration is implemented to gastroscope images and brain cryosection images. The experimental results show that our color ensemble registration can successfully align color ensemble images with stable color component's transformation.