Fast Texture Synthesis Based on Dynamic Space Partitioning

We deal with the high complexity problem of searching similar neighborhood in texture synthesis. By using nearest neighbor searching algorithm based on dynamic space partitioning, we propose a fast texture synthesis method. The input sample texture image can be constructed into searching space composed by multi-dimensional vectors, and we partition the space with establishing data structure and update the candidate list to find the most similar neighborhood. Experimental results show, compared with other typical accelerating algorithms such as TSVQ and PSO, our method is faster and has better visual synthesized results which are closer to the results of exhaustive searching.