usually, realignment is used to correct subject head motion within an image timeseries acquired from the same subject. Since you do not have timeseries data, no realignment will be necessary. Coregistration also is a spatial preprocessing step to align images (with different imaging contrast) of the same subject, e.g. PET and MR images of the same subject. It is not meant to align subject images with a template or another image from a different subject.
Prior to statistical group analysis you will need to spatially normalise your images. While it is highly recommended to use low-noise, high-resolution data (such as T1 MRI images) to estimate spatial normalisation, you can also try to normalise your PET data directly. If necessary, you can help the normalisation process by checking image orientation and position of the mm origin. Image orientation should match the orientation of the canonical or template images shipped with SPM, and the mm origin should be close to the AC. Use CheckReg to simultaneously display one of the SPM images and your PET data, and use "Reorient" from the CheckReg context menu to reorient each PET image roughly into position. Whether normalisation works reasonably well will depend a lot on the contrast and resolution of your PET images.
Hope this helps,