【作者】 覃庆炎；

【导师】 王绪本；

【作者基本信息】 成都理工大学 ， 信号与信息处理， 2008， 硕士

【摘要】 简牍作为中国古代重要的一种书写材料,保存了丰富的历史文化信息,是我国宝贵的文化财富。简牍图像增强是简牍文字修复和文字信息提取最为关键的一个步骤。由于大多数简牍实物长期埋存地下,出土时重叠、粘连,有的腐蚀、朽烂,文字信息出现脱落,墨迹扩散,肉眼根本难以识别简牍上的文字符号,同时在简牍数字图像获取过程中外界光照的非均匀分布,导致了大多数的简牍图像都存在着对比度差、分辨率低的缺点。简牍图像中暗区中包含了很重要的文字信息,由于这些信息的能见度较低,因而对简牍暗区的增强,提高暗区的可视度对简牍考古工作者来讲意义十分重大。传统的图像增强方法虽然能够使图像的对比度增强,但通常会丢失图像暗区的细节信息,并且会导致图像颜色的失真。所以需要一种有效的简牍图像增强算法,在提高图像对比度的同时,能有效地消除光照对简牍图像的退化影响从而增强图像暗区的细节信息并且使简牍图像具有较高的色彩保真度。本文以长沙简牍博物馆所收藏的简牍文物为实物模型,以该馆所提供的简牍数字图像为研究对象,深入研究简牍图像特点,在简牍图像增强处理的各阶段进行了大量实验,引入了一套适用于简牍图像增强处理算法:(1)多尺度Retinex算法Retinex是一种建立在科学实验和科学分析基础上的图像增强理论,其实质就是从图像中抛开照射分量的影响来获得物体的反射分量,即获得物体本来的面貌。与其它图像增强方法相比,Retinex算法具有锐化、颜色恒常性、动态范围压缩大、色彩保真度高等特点。本文对该理论进行了重点研究,并采用多尺度Retinex算法来消除光照对简牍图像的退化影响,以达到增强图像暗区细节信息和使简牍图像具有较好的色彩保真度的目的。(2)gain/offset修正通常利用多尺度Retinex算法处理后的图像像素值会出现负值,而且会超出显示器的显示范围的情况,这就需要将其平移和压缩到显示器显示围内,本文利用gain/offset方法对图像像素进行修正,再把修正后的图像的灰度值映射到显示器显示的灰度范围(0-255)。(3)对比度拉伸根据我们对几百张原始简牍的研究发现,这些简牍图像的直方图的灰度值通常处于20至220的区间,没有充分的利用整个灰度级(0-255),表现在简牍图像上就是对比度差,整个图像看起来比较黑。因此本文采用对比度拉伸算法来实现简牍图像高对比度的目的。试验结果表明:本文所采用的算法在提高简牍图像对比度的同时,能显著地增强图像暗区的细节信息并且使简牍图像具有较好较高的色彩保真度。该方法克服了目前常规的图像增强方法的不足,能够满足简牍考古人员的要求。更多还原

【Abstract】 Bamboo script preserves the rich historical and cultural information. As a writing material in ancient China,Bamboo script is the valuable cultural wealth. Enhancement of the Bamboo script Image is the most key procedure in character reconstruction and text extraction.As most of the Bamboo scripts were buried underground for a very long period,almost all of the Bamboo script the lap over together.They were conglutination and some of them even corrupted.The characters on the Bamboo scripts were fallen off and it is hard to identify. On the other hand,most of the Bamboo script were photoed under poor condition,therefor most of the Bamboo script were in the poor contrast condition. As the dark area in the image includes very importance detail about the character,so the enhancement of the dark area in the Bamboo script is the most useful for archaeologist.In this paper we made the Bamboo scripts from Changsha Jiandu museum as a model and taken the Bamboo script images as our research objects. This paper designs one kind of bamboo scripts images enhancement algorithm according to bamboo scripts image’s characteristics and achieves the following specific functional modules:(1)multi-scale Retinex(MSR) algorithmThe Retinex is one kind of image enhancement theory and it eatablished at the scientific experiment and the scientific analysis.It essence is to ’skinning’ the illumination in the original image and to yield the reflectance image,that is,the original appearance of the object.Compares with conventional enhancement methods,the Retinex has shown itself to be a very versatile automatic method and can simultaneously provide sharpening,color constancy,dynamic range compression and color rendition. Based on the theory of Retinex, the multi-scale Retinex algorithm is employed to remove the luminance degradation of the Bamboo script image and enhance the dark area and make the Bamboo script more color rendition. (2)gain/offset algorithmUsually after the multi-scale Retinex process,some of the image’s Pixels are negative and also will beyond the display’s ability,so we need to shift and compress to the display’s scope(0-255) .In this paper,the gain/offset algorithm is employed to achieve this aim.(3) contrast stretching algorithmBased on the research of hundred of Bamboo scripts,we made the conclusion that the histogram of the Bamboo scripts usually between 20 to 220.And not take full advantage of all of the gray level(0-255), the image looks very dark and not clear.So in this paper the simple contrast stretching algorithm is deployed to improve the output brightness.The results of experiment show that our algorithm can improve effectively not only the contrast but also can observably enhanced the information of hidden area in the Bamboo script image and provided better tonal rendition. The algorithm can overcome the lack of enhancement of traditional enhancement methods and satisfy the archaeologist’s demand.更多还原