Abstract
During the development of neuroimaging, numerous analyses were performed to identify population differences, such as studies on age, gender, and diseases. Researchers first normalized the brain image and then identified features that represent key differences between groups. In these studies, the question of whether normalization (a pre-processing step widely used in neuroimaging studies) reduces the diversity of brains was largely ignored. There are a few studies that identify the differences between individuals after normalization. In the current study, we analyzed brain diversity on an individual level, both qualitatively and quantitatively. The main idea was to utilize brain images for identity authentication. First, the brain images were normalized and registered. Then, a pixel-level matching method was developed to compute the identity difference between different images for matching. Finally, by analyzing the performance of the proposed brain recognition strategy, the individual differences in brain images were evaluated. Experimental results on a 150-subject database showed that the proposed approach could achieve a 100% identification ratio, which indicated distinct differences between individuals after normalization. Thus, the results proved that after the normalization stage, brain images retain their main distinguishing information and features. Based on this result, we suggest that diversity (individual differences) should be considered when conducting group analysis, and that this approach may facilitate group pattern classification.