基于优先权改进和块划分的图像修复

发布时间：2018-05-27 06:22

本文选题：图像修复 + 置信项　；参考：《中国图象图形学报》2017年09期

【摘要】：目的针对基于样本块的Criminisi图像修复算法易发生置信项迅速下降趋于零,使优先权计算公式失效,导致修复顺序错乱造成的修复效果失真问题,以及在搜索匹配块时存在的搜索范围过大,效率过低,易出现匹配到不符合视觉效果的纹理块问题,提出一种基于优先权改进和块划分的图像修复算法。方法首先重新定义优先权中的置信项,用样本块中的棋盘距离替代原计算公式,保证优先权一直发挥作用,从而减少因修复顺序不合理造成的错误匹配;其次根据图像纹理信息将其自适应划分为不同大小的图像块,使待修复样本块只在具有相似特征的图像块区域内搜索匹配。结果实验结果表明,新定义的优先权,保证了修复算法的正常进行,改善了修复图像的视觉效果;由图像自适应块划分引导匹配过程,可使匹配在更少的候选块中进行,提高了算法速度。将本文方法与3种全局搜索匹配方法和1种局部搜索匹配方法进行修复结果对比分析,本文方法的修复结果视觉完整性较好,而且修复时间小于其中3种算法。结论通过改进Criminisi算法优先权中的置信项,避免因其趋于零导致的修复顺序错乱造成的错误累积情况的发生;同时通过改进待修复匹配块的搜索范围,对整幅图像进行自适应块划分,使搜索只在相似块中进行,不仅减少了时间,而且提高了匹配的准确性。本文方法对于自然图像中大面积目标物体移除方面有较好的应用,可获得较满意的修复效果。
[Abstract]:Aim to solve the problem that Criminisi image restoration algorithm based on sample block is prone to rapidly decrease the confidence item to zero, make the priority calculation formula invalid, and lead to the distortion of the repair effect caused by the repair sequence disorder. And when searching matching blocks, the search range is too large, the efficiency is too low, and the texture blocks are easily matched to the visual effect. An image restoration algorithm based on priority improvement and block partitioning is proposed. Methods firstly, the confidence items in priority are redefined, and the original calculation formula is replaced by the chessboard distance in the sample block, so as to ensure that priority always plays a role, so as to reduce the mismatch caused by the unreasonable repair sequence. Secondly, according to the texture information of the image, it is adaptively divided into different size image blocks, so that the sample blocks to be repaired only search for matching in the image blocks with similar characteristics. Results the experimental results show that the new definition of priority ensures the normal implementation of the restoration algorithm and improves the visual effect of the restored image, and that the matching process can be guided by image adaptive block partitioning, so that the matching can be carried out in fewer candidate blocks. The algorithm speed is improved. Compared with three global search matching methods and one local search matching method, the results obtained in this paper have better visual integrity, and the repair time is less than that of the three algorithms. Conclusion by improving the confidence items in the priority of the Criminisi algorithm, we can avoid the accumulation of errors caused by the repair order disorder caused by its tendency to zero, and improve the search range of the matching blocks to be repaired. The adaptive block partition of the whole image makes the search only in the similar block, which not only reduces the time, but also improves the accuracy of matching. This method has a good application in the removal of large area objects in natural images and can obtain satisfactory results.
【作者单位】：南昌航空大学计算机视觉研究所;
【基金】：国家自然科学基金项目(61165011,61662049)~~
【分类号】：TP391.41

【相似文献】