Pradeep,
Regarding FWHM estimates for group analysis, let me quote the following paragraph from website
http://afni.nimh.nih.gov/sscc/gangc/mcc.html:
(3) Obtain spatial correlation (smoothness)
If you used 3dBlurToFWHM during preprocessing for all subjects, then simply adopt the numbers (under option -FWHM or -FWHMxy) from there. If you didn't, obtain the residual time series for each subject (through -errts in 3dDeconvolve), and then use 3dFHWMx on the residual time series in EPI resolution:
3dFWHMx -dset errts+orig -mask ... -out _fwhm.txt
Average the numbers (they should not vary much) across subjects. Using the input of 3dDeconvolve may work well too in the above command line.
Gang