【摘要】 针对Criminisi图像修复算法中样本块选择的随意性,导致图像修复效果欠佳的缺陷,提出一种针对纹理图像的Criminisi改进算法。该算法通过对纹理图像预处理后获得的二值化图像进行统计分析,获得图像的行和列纹理曲线,进而得到纹理曲线的极值点变化曲线,并最终获得样本块的近似尺寸。仿真实验表明,将本文获取的样本尺寸作为Criminisi修复算法的样本尺寸,获得了较传统Criminisi算法更好的视觉效果。更多还原

【Abstract】 Because of the casualness of exemplar patch selection, Criminisi algorithm leads to the effect of image completion notwell. To avoiding it, this paper put forward an improved algorithm for Criminisi according to texture image. By analyzing the binary image obtained through preprocessing texture image, itgets row and column texture curve. Furthermore, itgets the extreme pointcurve.Finally,itgets approximate size of exemplarpatch.Simulation results show that the improve algorithm can obtain better visual effect than traditional Criminisi.更多还原