Abstract: The traditional examplar-based texture synthesis algorithm is lack of adaptability since the Size Of Patches（SOP） needs to be defined artificially in advance or adjusted according to the completion effect. In order to solve this problem, a method for adaptively selecting the SOP is proposed. The images are pre-processed with the image decomposition technique. The image texture features are extracted. The SOP is automatically selected according to the relationship with the extracted texture features. Simulation results show that the suitable effect of texture synthesis can be obtained with the SOP selected by the proposed method.

Key words: image completion, texture synthesis, texture feature, co-occurrence matrix, Size Of Patches（SOP）

摘要： 在传统基于样本块的纹理合成算法中，样本块的大小需要靠人工设定，或根据图像修复效果进行调节，缺乏自适应性。为解决上述问题，通过提取图像的相关纹理特征，提出一种样本块大小自适应选取方法。利用图像分解技术对图像进行预处理，利用图像灰度共生矩阵提取待修复图像的纹理特征，通过确定相关纹理特征与最优样本块大小的相互关系，实现了样本块大小的自动选取。仿真实验结果表明，利用文中方法自适应选取样本块大小，对自然图像能够获得理想的纹理合成效果。

关键词: 图像修复, 纹理合成, 纹理特征, 灰度共生矩阵, 样本块大小