激光与光电子学进展, 2020, 57 (12): 121020, 网络出版: 2020-06-03
基于信息熵和结构特性的敦煌壁画修复算法 下载: 1007次
Dunhuang Mural Inpainting Algorithm Based on Information Entropy and Structural Characteristics
图像处理 壁画修复 信息熵 优先权模型 Criminisi算法 image processing mural inpainting information entropy priority model Criminisi algorithm
摘要
Criminisi图像修复算法优先权计算中结构信息考虑不足和匹配时仅依靠颜色距离选择,致使壁画修复过程中易出现结构传播错误和像素错误匹配,鉴于此,提出了一种基于信息熵和结构特性的壁画修复算法。先在计算优先权函数时,引入度量像素块复杂度的信息熵,改进的优先权函数确定了最优待修补块,使结构信息丰富的区域优先修复;再采用样本颜色特征和块间协方差结合的方式确定匹配块集合,根据块间欧氏距离确定最佳匹配块;最后通过迭代更新完成壁画修复。对破损敦煌壁画进行实验,实验结果表明:该算法较好地克服了Criminisi算法错误匹配填充的问题,修复后获得了较好的视觉效果,提高了图像峰值信噪比等客观评价值。
Abstract
In view of the insufficient consideration of structural information in the priority calculation of the Criminisi image inpainting algorithm and the fact that matching only relies on the color distance selection, the mural repair process is prone to structural propagation errors and pixel mismatches. To address this, a mural inpainting algorithm based on information entropy and structural characteristics is proposed in this study. First, when calculating the priority function, the information entropy of measuring the complexity of the pixel block is introduced, and the optimal block to be repaired is determined by improving the priority function to preferentially repair the regions with rich structural information. Then, the matching block is determined by combining the sample color feature and the covariance similarity between blocks, and then the best matching block is determined through the Euclidean distance between the blocks. Finally, the mural inpainting is completed through iterative updating. Experiments on damaged Dunhuang murals show that the proposed algorithm overcomes the problem of the Criminisi algorithm mismatching and filling. Subsequent to the repair, good visual effects are obtained, and objective evaluation values such as peak signal-to-noise ratio of the image are improved.
陈永, 艾亚鹏, 陈锦. 基于信息熵和结构特性的敦煌壁画修复算法[J]. 激光与光电子学进展, 2020, 57(12): 121020. Yong Chen, Yapeng Ai, Jin Chen. Dunhuang Mural Inpainting Algorithm Based on Information Entropy and Structural Characteristics[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121020.