基于结构分量和信息熵的Criminisi图像修复算法

针对Criminisi图像修复算法中优先级计算易受 图像纹理影响的问题,提出了改 进的基于图像结构分量的优先级函数。首先采用变分分解模型,将待修补图像分解为结构分 量和 纹理分量；其次基于结构分量计算数据项,排除纹理的影响；然后在优先权函数中 引入度量像素块复杂度的信息熵,将像素块中除了中心点之外其它位置的结构信息 融 入到优先权的计算中,使修补次序进一步向结构丰富的像素块倾斜；最后将优先权函数 表 示为置信度、数据项和信息熵的加权和,以解决传统Criminisi算法优先权随着置信度 迅速 下降为零而造成修复次序出现偏差的不足。新的优先权函数排除了像素块中在计算数据项时 纹 理的影响,并且融合更多的结构信息,使修复次序更加准确。实验结果表明,对于 不 同的人工图像和自然图像,本文模型都能取得较为满意的修复结果。

An improved Criminisi image inpainting algorithm based on structure component an d information entropy

In order to avoid the influence caused by image texture in priority computabion of Criminisi image inpainting algorithm,we propose an improved priority function based on im age structure component in this paper.Firstly,the damaged image is decomposed into the structu re component and the texture component by a variational decomposition model.And then,we compute the data term based on structure component,which can avoid the influence caused by image texture in data term computabion.Moreover,the information entropy is introduced in the prio rity function to measure the complexity of the image block.In this way,not only the isophote cr oss the central point, but also the structure information at other positions are taken into account in priority computation, which further emphasizes the structure information in image inpainting.At last, the priority function is defined as a weighted sum of the confidence term,the data term and the infor mation entropy, which solves the drawback of the chaotic inpainting caused by the rapid decrease of pr iority to zero with the confidence.The new priority function avoids the influenc e of image texture,and combines more structure information.The experimental results on both noisy synthetic and real images demonstrate the effectiveness of the proposed algorithm.