结构相似度索引耦合最优稀疏表示的大规模损坏图像动态修复

当前的图像修复算法在处理小面积损坏图像时,可取得较佳的视觉质量;但难以用于高对比度边缘和高频分量的大面积损坏图像的修复,存在明显的模糊效应与块效应,使得修复质量较差。对此,提出了结构相似度索引耦合优化稀疏表示的大规模损坏图像动态修复机制。基于数据度与置信度构造图像块先验模型,提取损坏图像的已知块;再引入K-SVD算法和拉格朗日乘数机制,构造最优字典矩阵,优化稀疏表示,以重构目标图像损坏区域;并设计结构相似度索引与动态更新机制,估算稀疏表示系数,并动态更新字典矩阵,输出修复图像。最后测试了其机制性能。结果表明:与当前图像修复算法相比,在大规模损坏图像与高对比度边缘图像修复中,具有更好的修复效果、更高的相似度,显著消除了模糊效应与块效应。

The Study on the Large-Scale Damage Image Inpainting Mechanism Based on Structural Similarity Index Coupled Optimal Sparse Representations

the good visual effects can be got by the inpainting the small damage based on the current image inpainting algorithm; but these algorithm can not be applied to the image with high contrast edges and high frequency components, the obvious blur effect and blocking effect were existed to reduce the inpainting quality. So the large-scale damage image inpainting mechanism based on structural similarity index coupled optimal sparse representations. The optimal dictionary matrix was constructed to optimize the sparse representation for reconstructing the damage region of image by introducing the K-SVD algorithm and Lagrange multiplier mechanism; and the structural similarity index was designed to estimate the sparse representation coefficients and dynamic update the dictionary matrix to finish the image inpainting. Finally, the performance of this mechanism was tested. The results showed that: comparing with other image inpainting mechanism, the effect of image inpainting of this mechanism was best, and it can obviously eliminate the blur effect and blocking under the condition of large scale damage image inpainting.