摘要
针对基于全变分(TV)小波域图像修复算法,提出了一种新的数值计算方法。算法充分利用了图像待修复像素点周围八邻域的信息计算曲率在像素域的近似值。与传统近似求解方法不同,该数值近似计算方法不仅保证了近似结果具有更高的精确性,而且对噪声的鲁棒性好。对不同丢失率的图像用所提出的数值计算方法取得了较好的修复效果,尤其当小波系数丢失率较高时效果更为明显。所提方法可以为图像压缩导致的系数缺损修复提供解决思路。
This paper proposed a new numerical method for Total Variation （TV） wavelet inpainting algorithm. The information of neighborhood of damaged pixel was employed sufficiently to calculate curvature in pixel domain. In comparison with the traditional approximation solution method, the proposed algorithm has higher accuracy; moreover, the result is also less sensitive to noise. The experimental results which use different loss ratio of image show that the proposed method achieves better inpainting effect, especially when the wavelet coefficient is relatively high. The proposed method provides new idea for restoring the missing information in the wavelet domain.
出处
《计算机应用》
CSCD
北大核心
2013年第1期28-30,共3页
journal of Computer Applications
基金
国家自然科学基金资助项目(61162021)
西北民族大学科研创新团队资助项目
中央高校基本科研业务费专项资助项目(zyz2012078)
关键词
图像修复
小波域
全变分模型
数值计算
image inpainting
wavelet domain
Total Variation （TV） model
numerical calculation
