Nonrigid color image registration by minimizing residual complexity under illumination variations

In the presence of spatially-varying illumination changes, many intensity-based similarity measures cannot capture the complex interactions among the pixel intensities and result in wrong registration. Residual complexity (RC) measure can solve this problem and provide accurate and robust registration results in the same modality images registration. In this study, RC measure is generalized to nonrigid color image registration and the corresponding color image registration scheme is proposed. By nonrigid registration of three kinds of color images, four similarity measures, i.e. RC, mutual information, correlation coefficient and sum-of-squared differences, are contrasted and studied. The results of tests show that RC measure outperforms other three measures in the presence of illumination variations and geometrical transformation of color images.