摘要
在基于纹理合成的图像修复算法中,最佳样本块匹配算法存在匹配精度不高和时间复杂度高等问题。针对上述问题,首先构造了块匹配算法,采用矩阵相似度来计算模板块与样本块之间的匹配度,以相对较粗的粒度初步选出最佳样本块的候选集。然后,又构造了像素点匹配算法,采用模板块与候选最佳样本块之间的误差矩阵的内积来计算对应像素点之间的匹配度,以更细的粒度来确定最终的最佳样本块。块匹配算法降低了时间复杂度,像素点匹配算法提高了匹配精度,因此,在此基础上构造的基于相似矩阵的最佳样本块匹配算法能够在不增加时间复杂度的情况下提高算法的匹配精度。实例验证结果表明,与当前基于纹理的图像修复算法相比,该算法的匹配精度提高,时间复杂度降低。
In the image inpainting algorithm based on texture synthesis, the matching accuracy of optimal exemplar matching is not high but its time complexity is too high,which eventually leads to errors in image inpainting. Focusing on these two issues, firstly, the block matching algorithm was constructed and the matching degree between template and exemplars was measured by using matrix similarity, so, the candidate set of optimal exemplar was preliminarily de- termined in relatively coarse granularity. Secondly, the pixel matching algorithm was constructed and the matching de- gree between corresponding pixels was measured by inner product of error matrix between template and candidate ex- emplar, so, the final optimal exemplar was determined in fine granularity. The block matching algorithm can reduce the time complexity while the pixel matching algorithm can improve matching accuracy, therefore, the optimal exemplar matching algorithm based on matrix similarity can improve matching accuracy without raising time complexity. The ex- perimental results demonstrate that, compared with current texture-based inpainting algorithm, the proposed algorithm can improve matching accuracy and reduce time complexity.
出处
《计算机科学》
CSCD
北大核心
2014年第1期307-310,共4页
Computer Science
基金
国家社会科学基金项目(12EF119)
国家级大学生创新创业训练计划项目(201210 694019)资助
关键词
图像修复
块匹配
矩阵相似度
像素点匹配
Image inpainting, Block matching, Matrix similarity, Pixel matching