基于结构分量和信息熵的Criminisi图像修复算法

发布时间：2018-02-02 01:53

本文关键词： 图像修复优先级函数变分图像分解结构分量信息熵纹理　出处：《光电子·激光》2017年01期 　论文类型：期刊论文

【摘要】：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。
[Abstract]:In order to solve the problem that priority calculation is easily affected by image texture in Criminisi image restoration algorithm, an improved priority function based on image structure component is proposed. First, the variational decomposition model is adopted. Decomposing the image to be repaired into structural component and texture component; Secondly, the data items are calculated based on the structural components to eliminate the influence of texture. Then the information entropy which measures the complexity of pixel block is introduced into the priority function, and the structure information of the pixel block other than the center point is incorporated into the calculation of priority. The patching order is further tilted to the rich pixel blocks; Finally, the priority function is expressed as the weighted sum of confidence, data item and information entropy. In order to solve the problem that the priority of the traditional Criminisi algorithm decreases rapidly to zero with the confidence degree falling to zero, the new priority function eliminates the shadow of the texture in the pixel block when calculating the data item. Ring. The experimental results show that the proposed model can obtain satisfactory results for different artificial and natural images.
【作者单位】：湖北民族学院理学院;
【基金】：国家自然科学基金(61561019) 国家科技支撑计划(2015BAK27B03) 湖北省自然科学基金(2015CFB262) 湖北民族学院博士启动基金(MY2015B001) 湖北省大学生创新创业训练计划(201510517004)资助项目
【分类号】：TP391.41
【正文快照】： 1引言在图像的获取、储存和传输过程中,有很多因素会引起图像上局部信息的缺损。为了保证图像信息的完整性,需要对受损图像进行填充修复[1]。近年出现了许多卓有成效的数字图像修复模型或方法。这些方法大体上可分为两大类[2,3]:一类是针对小尺度缺损的基于结构的图像修复技术

【相似文献】