In principle, the proposed approach assumes translation invariant homogeneity of a texture and is based on a generic MGRF model specified by a joint probability distribution on the GLCHs. As global statistics of image signal co-occurrences, the GLCHs are more sensitive to the spatial periodicity than to the local spatial variation of image signals. This leads to inefficient modelling and synthesising a class of weakly-homogeneous textures which exhibit both the global periodicity and the local deformation in image signals. Experimental results show that the proposed approach only effectively detects and reproduces the regular periodicity structure but neglects local deformation in most weakly-homogeneous textures. This is the major limitation of the present work. Figure 8.1 displays an difference image that results from subtracting the homogenised synthetic texture (centre) from the original texture D3 (left), which contains the missing part that the proposed approach fails to model.