基于主元分析法的图像修复

Image Inpainting Based on PCA

图像修复是图像处理、计算机视觉的热点研究领域,本文针对块状信息缺失的图像提出了一种基于统计意义的修复方法。根据包含待修复块的扩展区域确定样本集的范围,再利用主元分析法对样本集中对应的两组子区域进行统计分析,分别求出其特征向量作为新的坐标空间,最后根据扩展块中未丢失信息的区域部分在坐标空间中的投影,计算出待修复块中的值。实验结果显示该方法的修补效果较好,且简单易行,具有一定的实用性。

Image inpainting is one of the important research aspects in image processing and computer vision. A statistical learning method is presented to repair the block of missing image information. The sample blocks are located according to extended block in which the block of missing information is included, then two sets of eignvectors are obtained from sample blocks by using Principal Component Analysis, with coordinate values in eignvector space based on sample blocks and eignvectors reduced by sub sample blocks, the block of missing information can be repaired. Experimental results demonstrate a better visual effect can be achieved, and the method is easy to be realized, therefore it has a better practicability.