high performance computing on graphics processing units: hgpu.org

This paper presents a new algorithm for synthesising image texture. Texture synthesis is an important process in image post-production. Previous approaches can be classified as either parametric or nonparametric. Of these nonparametric approaches have achieved the most impressive results. Unfortunately, these methods generally suffer from high computational cost and difficulty in handling scale in the synthesis process. By introducing the wavelet decomposition as a basis for nonparametric texture synthesis, the advantages of nonparametric methods are combined with the computational efficiency and scale information readily available from the transform. The results show an order of magnitude improvement in computational speed and a better approximation of the dominant scale in the synthesised texture. The paper also presents a fast block searching algorithm using commodity graphics hardware. The computationally demanding neighbourhood search required by the Efros texture synthesis algorithm is implemented on the GPU. This results in an Efros texture synthesis algorithm which is up to 8 times faster than a similar C++ implementation.