基于MAP的单帧字符图像超分辨率重建  

MAP-Based Single-Frame Super-Resolution Reconstruction for Character Image

作　　者：李展[1,2] 陈清亮[1] 彭青玉[1,2] 张庆丰[1,2] 李伟祥[1] 

机构地区：[1]暨南大学计算机科学系,广东广州510632 [2]暨南大学天体测量动力学和空间科学研究中法联合实验室,广东广州510632

出　　处：《电子学报》2015年第1期191-197,共7页Acta Electronica Sinica

基　　金：国家自然科学基金(No.11273014;No.11403008;No.U1431227);中央高校基本科研业务费专项资金(No.21612413;No.21612414);广东省自然科学基金(No.10151063201000002);国家大学生创新创业训练计划项目(No.201310559044)

摘　　要：字符是很多图像处理应用关注的区域,增强字符图像分辨率往往能提高字符识别率.针对字符图像纵横和对角线方向纹理特征,提出了一种提取不同方向纹理及图像平滑度度量的新方法,并在最大后验概率框架下,实现了单帧图像超分辨率重建算法.算法利用自适应可调模板作为卷积核,将纹理信息引入图像先验模型,最终将超分辨率重建转化为一个目标函数最优化问题.实验表明,该算法能有效提高字符图像的识别率,且对噪声具有较好的鲁棒性.Characters are concerned in many image processing applications. Enhance resolutions of character images frequently lead to higher recognition rate. Considering vertical,horizontal and diagonal texture features of character images,a new image smoothness measurement as w ell as a new method for extracting image texture in different directions is proposed,and a single-frame super-resolution（ SR） reconstruction method is implemented under the maximum a posteriori（ MAP） framework. Using a self-adaptive flexible template as a convolution kernel,texture information is introduced into the image prior model. Therefore,SR reconstruction is finally converted into an objective function optimization problem. Experimental results show that the proposed algorithm effectively increases the recognition rate for character images and has preferable robustness to noise.

关 键 词：图像重建 超分辨率 最大后验估计 纹理分析 字符识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]