一种改进的基于样例的图像修复方法

AN IMPROVED EXEMPLAR-BASED METHOD FOR IMAGE INPAINTING

以基于样例修补的目标移除方法为基础,改进了基于样本块的图像修补方法。在自然图像上通过傅里叶变换,发现许多自然图像具有方向性纹理的特征,将搜索空间约束到纹理方向的范围,优化了Criminisi方法优先块的选取,提高了搜索精度;并通过在源图像区使用图像分割的方法实现分区,使搜索目标块仅在其相邻的源区域内搜索,进一步缩小样本图搜索范围,增强搜索的准确性。在自然图像上进行的实验结果表明:改进的方法不仅显著提高了图像修补的时间,且有效地维持了图像的线性结构,取得了良好的修补效果。

Sample block-based inpainting algorithm is improved on the basis of object removal method based on exemplar inpainting. It is found that a lot of nature images have directional texture feature with performing the Fourier transform on them. To constrain the search space to texture orientation scope can optimise the selection of priority block in Criminisi＇ s method and improve the search precision. Furthermore, an image segmentation method is used to partition in source image region to have the search objective block just search within its adjacent source region so as to further narrow the search area of the sample graph for improving the search precision. Experimental results made on nature images show that the improved method can greatly accelerate the inpainting time, and reserve the linearity structure of the image effectively as well as achieve preferable inpainting effect.