Multimodal image registration using mean and variance of joint intensity distribution

Mean and variance are used as two of several import descriptors of distribution in probability theory and statistics. In this paper, we present several new intensity-based multimodal image registration methods using mean and variance of data. Partition intensity uniformity (PIU) is the first successful medical multimodal image registration algorithm. PIU method will be ill-conditioned when the mean of intensity of voxels, which positions in one image are correspond to given intensity value in another image, is very small. A new adjustable parameter alpha is introduced into PIU. What is more, mean parameter that refers to as expectation is removed off from PIU function, then two new different PIU measures are proposed. These new similarity measures are preliminarily validated by multi-modal medical image data registering experiments from calculating time, robustness to noise and influence of image window overlapping region. The results of tests show the new modified PIU metric may have better performance.