Frequency modulated thermography is becoming a popular thermal non-destructive testing (TNDT) technique which like other active thermographic techniques, requires an external heating stimulus, preferably on a blackened surface. It is however, not immune to non-ideal situation like non-uniform heating and surface emissivity variation. The phase image helps to reduce the effect of surface emissivity variation to some extent, but is inadequate in case of large variations. Further, structural noise can significantly interfere with the process of defect detection, as in the case of carbon fibre composite materials. This paper proposes two image reconstruction algorithms for off-line reduction of structural noise. The first utilizes the periodic nature of structural noise if present, and removes it using 2-dimensional Fourier transformation and a spatial band-stop filter. The other uses time serial reconstruction algorithm to remove such noise. The latter was originally proposed by the authors to remove artifacts from surface emissivity variation on metallic samples. The performances of both algorithms are successfully demonstrated on a carbon fibre test piece, having 2mm, 4mm, and 6mm diameter back drilled holes at various depths ranging from 0.25mm to 2mm in steps of 0.25mm.