Multiresolution texture synthesis using real-time pattern matching

This paper presents a novel pixel-based multiresolution texture synthesis algorithm. First, pyramids are constructed for the sample and output separately, and a novel method is designed to synthesize the lowest resolution texture so that a real-time pattern matching algorithm can be used to expedite the neighborhood matching. Then, to sweep out the artifacts and improve the synthesis quality, iterative refinement to the lowest resolution texture is done. At lasts a new multiresolution synthesis algorithm is developed to synthesize other images in the output pyramid. Given the lower resolution image, the higher resolution image can be obtained in real time with little loss in quality. These three features contribute to the outstanding result that textures with satisfactory quality can be synthesized in seconds. Experimental results show the effectiveness of the proposed algorithm.