Estimating the mean value of a stationary spatial process from one finite realisation is possible if a representative sample can be found. Nevertheless, the question of the uncertainty quickly arises. In this paper, images are considered as collections of randomly distributed objects. In this context, we employ the point process formalism and the pair correlation function to estimate the variance of the objects mean surface ratio. Our approach is evaluated on two theoretical models. Finally, the applicability of the method on real images is discussed and results are given on composite material images.