A non-parametric texture synthesis technique is applied to the problem of digital image inpainting. The technique described primarily handles homogeneous texture images. Based on the Castellanos-Williams algorithm, the method implicitly assumes a Markov random field model for textured image regions. The non-parametric sampling procedure for image inpainting utilizes a series of binary mask Gaussian pyramid level textures. Texture is synthesized in a coarse-to-fine order. The Laplacian pyramid transform is used in the implementation. Resolution hierarchy plays a critical role in the analysis, synthesis and sampling process for efficiency and flexibility. The degree of randomness is crucial in the sampling routine. Here, the sampling process randomly and uniquely chooses a pixel from the initial guess level, combining with the sampling with replacement policy. Therefore, the sampling procedure generates a distribution that is very similar to the sample by running more than one time and taking into account the neighborhood adjusting factor, k. The combination of these contributes achievements to fulfill the best performance of image inpainting.