Fast Texture Synthesis Using Feature Matching

A novel tile-based algorithm is proposed to achieve real-time synthesis of structural texture. The algorithm operates by the joint use of tiling and feature matching. Firstly, quantitative evaluation of texture regularity is introduced to classify different texture. Secondly, a new geometry-color measure which evaluates texture similarity from both geometry and color distance is brought forward to pick the best sample pattern from input texture. Thirdly, in contrast to traditional tile-based algorithms that choose the cutting curve by color assess, ours uses feature matching and feature deformation to reduce the appearance of conspicuous seams and dislocated features. This operation ensures the quality of the whole tiling process while guaranteeing the synthesis rate. Experimental results verify that the proposed algorithm is not only more efficient than current technology using feature matching but also can achieve better results than traditional tile-based algorithm.