邻域窗口权重变分的图像修复

目的 传统的基于样本块的图像修复算法对于破损区域周围既含有几何结构信息又含有丰富纹理信息的情形,修复过程中易出现纹理延伸现象和错误样本块问题,该研究旨在改进传统的修复算法,提出基于邻域窗口权重变分的图像修复算法。方法 该算法利用领域窗口总变分和内在变分构造出权重变分,通过对Criminisi算法中的优先级测度进行加权,提高了对几何结构信息和纹理信息的辨识能力,使几何结构信息得到优先修复;同时,在像素块的匹配过程中,通过引入整体结构差异算子,并与传统的颜色匹配相结合,提高了匹配精度。结果 改进的算法很好地克服了原算法中的纹理延伸和误匹配问题,保持了修复结果的视觉连通性,其峰值信噪比相比原算法提高23 dB。结论 相比于Criminisi算法及其相应的改进算法,本文算法能够对既含有几何结构又含有丰富纹理信息的破损区域取得更好的修复效果,同时,也能高效修复一般的破损区域,从而具有更好的普适性。

Image inpainting with weight variation of neighborhood window

Objective This study aims to overcome the challenges of traditional image inpainting algorithms, wherein texture extension may occur and some incorrect samples may be selected as candidate patches when sample-based algorithms are employed to inpaint the damaged region with complex geometric structure and rich texture. Method An image inpainting method based on weight variation of neighborhood window is proposed. In the method, so-called weight variation is introduced by combining total variation and intrinsic variation in a neighborhood window to modify the priority measure in Criminisi's algorithm. With the proposed method, the ability of identifying geometric and texture information has been improved and geometric information can be urgently inpainted. Meanwhile, matching accuracy has been improved by introducing structure difference operator in combination with pixel color comparison. Result Compared with other recent algorithm, the proposed algorithm can settle the problem of texture expansion and block mismatching and can maintain visual connectivity. Moreover, the peak signal-to-noise ratio(PSNR) of its inpainting result is improved by 2 dB to 3 dB. Conclusion Compared with the original Criminisi's algorithm and its improved algorithms, the proposed algorithm can achieve better result in inpainting the damaged region with both geometric structure and rich texture, as well as in inpainting some ordinary damaged region. Thus, the proposed algorithm has generality.