基于参数自适应调整的退化藏文古籍图像二值化算法研究

In order to better reveal the contents recorded in the Tibetan ancient document, we can segment the text from ancient document image use the binarization algorithm. As show in the experimental result of reference files, a problem associated with all the proposed binarization algorithms is that they cannot deal with all types of degradation and different datasets. In order to solve the problems that arise in the binarization of Tibetan old document images characterized by many types of degradation, including poor quality, variations in image contrast and blurred letters, the study aims to design a Tibetan ancient document image binarization algorithm, parameter adaptive adjusting technique is used for this algorithm to achieve the better results. The study of degraded Tibetan ancient document image binarization is important for the ensuing document image processing tasks such as document digitization and full text retrieval and provides a huge potential market for application. To achieve the desired goal, we will address the following questions: how to construct local contrast image which can highlight the text message, how to estimate Tibetan ancient image background using the image inpainting algorithm, how to adaptive compensate the image contrast based on the estimated document background, how to design a binarization algorithm which combining several complementary binarization techniques, how to achieve parameter adaptive adjusting technique through the Racing algorithms.

为更好地揭示藏文古籍所记载内容，常常采用二值化算法将退化古籍图像中的文本和背景分割开。参考文献的实验结果表明，众多的二值化算法中没有一个能够处理所有的古籍退化类型及所有的古籍图像数据库。本课题针对藏文古籍图像二值化时存在的质量差、对比度渐变及字迹模糊等问题展开研究，拟设计一种能够根据藏文古籍图像退化情况自适应调节参数的二值化算法，以期得到满意的处理结果。该研究是古籍数字化和全文检索的必要步骤，蕴藏着巨大的应用价值。为实现预期目标，本课题计划解决以下几个问题：如何构造一种突出文字信息的局部对比度图像；如何用图像背景修复技术来估计古籍图像背景噪声；如何利用图像背景噪声来自适应补偿图像对比度；如何融合多个互补优缺点的二值化算法，并做相应的改进使其适合退化藏文古籍；如何将竞赛算法应用到所设计的二值化算法，从而实现参数自适应选择。

为了解决藏文古籍存在的质量差、对比度低及字迹模糊等退化问题对二值化的影响，本项目重点研究了基于参数自适应调整的退化藏文古籍图像二值化算法。主要研究内容如下：第一，通过整合高斯平滑降噪和Kovesi相位保持降噪算法以及归一化局部最大值最小值等方法，成功构造了一种突出文字信息的局部对比度图像，并设计了一种古籍背景修复模板计算方法；第二，通过Ntirogiannis图像修复算法、形态学闭操作、局部图像梯度以及Canny边缘检测算法，并结合所计算背景修复模板，实现了古籍背景噪声估计算法；第三，针对古籍图像存在的不同退化类型，利用所估计古籍背景噪声实现了图像对比度自适应补偿算法，并用Howe提出的基于拉普拉斯能量的二值化算法对补偿后的古籍进行二值化；第四，融合了多种二值化算法的优缺点，构造了一种新的拉普拉斯( Laplacian) 能量和古籍图像局部均值计算方法来改进Howe算法的性能；第五，设计了一种基于迭代弗里德曼竞赛算法（I/F-Race）的离线参数调整方法来自适应调节所设计二值化算法的参数配置。采用DIBCO以及本项目制作的藏文古籍测试数据集对所提出的算法进行测试，取得了较好的实验结果。该算法的深入研究，将对古籍图像二值化的进一步发展和完善具有重要的理论和应用价值，为藏文古籍数字化和全文检索奠定基础。