| 摘要: |
| 文字作为图书封面中的重要组成部分,包含丰富的语义信息.从复杂彩色图像中准确地获取文本信息,并结合现有的图像检索技术,可以进一步提高图书检索的精确度.针对中文图书封面文本的特点,采用基于连通分量的方法定位文本区域.首先通过颜色聚类将图像其分解为一系列的二值图像,然后依汉字结构合并各个图像中的连通分量,生成候选文本区域;通过文本验证进一步滤除非文本区域.定位获得的文本区域作为图书封面的显著区域,对其提取Hu不变矩特征用于图像匹配.经实验证实,该方法取得了较好的检索效果,表明了文本信息对于图书检索的重要性. |
| 关键词: 文本定位 连通分量 颜色聚类 Hu不变矩 |
| DOI： |
| 分类号: |
| 基金项目:山东省自然科学基金(ZR2009GL014); 山东省中青年科学家奖励基金(BS2010DX037); 文化部科技创新基金(46-2010); 中央高校基本科研基金(09CX04044A, 10CX04043A,10CX04014B, 11CX04053A, 11CX06086A,12CX06083A, 12CX06086A) |
|
| Chinese Book Cover Text Location and Chinese Book Retrieval |
|
LIU Yu-Jie1, LI Feng1, LI Zong-Min1, LI Hua2, LIN Mao3
|
|
1.College of Computer and Communication Engineering, China University of Petroleum, Qingdao 266580, China;2.Institute of Computing Technology, The Chinese Academy of Sciences, Beijing 100190, China;3.Research Institute of Exploration and Development, Xinjiang Oilfield, Urumqi 830013, China
|
| Abstract: |
| As an important part of book covers, characters contain rich semantic information. By extracting accurate information from complex color images, and combining it with content-based image retrieval technology, it is possible to further improve the accuracy of book retrieval. According to the characteristics of text information in Chinese book covers, this paper proposes connected components methods to locate the text regions. At first, the grayscale image is decomposed to a series of binary images and merged to connect components in each image, according to the structures of Chinese characters, generating candidate text regions. Additionally, text verification is used to rule out non-text regions. The result regions are regarded as the prominent regions of book cover, further this paper use Hu moment invariant to extract features for image matching. Experiments show the results of this method are fairly good, proving the importance of text information to book retrieval. |
| Key words: text location connected component color clustering Hu moment invariant |
