摘要
仿效人类的视觉认知过程,提出面向目标的图像超分辨率算法.只需从一幅车牌图像就可以恢复目标的细节信息.该算法使用先检测、后重建的思路,通过联合稀疏编码建立目标高低分辨率图像片之间的关系,以目标可以稀疏表示为先验,检测到目标区域后,通过压缩感知重建图像.实验表明,重建图像的峰值信噪比(PSNR)较传统方法约有2 dB的改善.此外,还验证了超分辨率重建改善了车牌识别结果,可以消除20%的错误识别字符.
An object-oriented image super-resolution approach is proposed, which imitates visual cognition of human beings. The reconstruction procedure needs only a single image of the license plate. In the training stage, the corresponding relationship between high and low image patches is built by combined sparse coding. After an object is detected, the low resolution object image is reconstructed by compressing sensing under assumption of sparse representation. Experimental results on license plate images show that the PSNR is improved by nearly 2 dB compared to conventional non-object oriented strategy, and 20% of misrecognized characters are correctly recognized after reconstruction.
出处
《中国科学院研究生院学报》
CAS
CSCD
北大核心
2013年第1期137-143,共7页
Journal of the Graduate School of the Chinese Academy of Sciences
基金
中国科学院百人计划(99T300CEA2)
国家科技重大专项(2010ZX03006-001-02)
国家高技术研究发展计划(2009AA12Z143)资助
关键词
面向目标
超分辨率
压缩感知
稀疏编码
邻接特征
object oriented
super-resolution
compressed sensing
sparse coding
neighborfeature
