A Fast Texture Matching Algorithm Based on Multi-resolution and Partition

Texture matching, that directly affects computational costs and the quality of inpainting, is one of the most important steps in texture synthesis applied to image inpainting. To obtain the global optimal solution efficiently, the multi-resolution and equal image partition is introduced in this paper. According to the resolution of input image, we build images with multi-resolution. Then all the images are partitioned into blocks equally and the proposed texture matching algorithm is block-based one. Based on the images with lower resolutions, the blocks with the most matching patches of lower sum of absolute differences (SAD) value are eliminated gradually. Performance evaluation is taken to the object removal. The experiments show that the inpainting results obtained by the proposed texture matching method are visually similar with the ones obtained by full search (FS) algorithm, while the inpainting algorithm with the proposed texture matching method is much more efficient than the one with FS method.