Diffuse optical tomography (DOT) for brain imaging has the potential to be an alternative human brain mapping technique when MRI imaging is not applicable. It recovers tissue chromophore concentrations of brain tissue through measures of light transmission to monitor for example the resting-state brain dynamics. This imaging technique relies on simulation of the light propagation which can be generated based on a subject-specific model. There has been some study on using rigid atlas models as alternatives for model based DOT when subject-specific anatomical data is not available; but there is still a lack of detailed analysis between geometrical accuracy and internal light propagation in tissue for atlas-based DOT. This work is focused on High-Density DOT (HD-DOT) of the whole cortex based on atlas models from 11 different rigid registration algorithms across 24 subjects, and the results are evaluated in 19 areas of the human head. The correlation between geometrical surface error and internal light propagation errors is strong in most area but varies in different regions from R2 = 0.74 in the region around top of the head to R2 = 0.98 in the region around the temples. In the 11 registration methods, basic-4-landmark registration with 4.2mm average surface error and 50% average internal light propagation errors is shown to be the least accurate registration method whereas full-head landmark with non-iterative point to point with 1.7mm average surface error and 32% average internal light propagation error is shown to be the most accurate registration method for atlas-based DOT.