关于图像块分类稀疏表示的超分辨率重构算法.pdfVIP

5 Vo .40 No . 5 20 12 5 ACTA ELECTRONICA SINICA May 20 12 练秋生, 张 伟 ( , 066004) : , . , . , . . , . , , , . : ; ; ; : TN91173 : A : 0372- 2112 ( 2012) 05- 0920-06 URL: http: / / www . ejourna . org. cn DOI : 10 . 3969/ j . issn . 0372- 2112 . 2012 . 05. 010 Image Supe-r Resolution Algorithms Based on Sparse Representation of Classified Image Patches LIAN Qiu- sheng, ZHANG Wei ( I nstitute of I nf ormati on Sci ence and Technol ogy , Yanshan University , Qinhuangdao , H ebei 066004, China ) Abstract : At present , super-reso ution a gorithm s based on spar se representation of image patches exp o it sing e dictionary to represent the image patches, w hich can not ref ect the diff erences of various image patches types. In this paper, a nove method based on sparse representation of c assif ied image patches is proposed to overcome this disadvantage. In this method, image patches are f irst y divided into smooth patches, diff erent directiona edge patches and irregu ar structure patches by oca features. Then these c assif ied patches are app ied into training the corresponding high and ow reso ution dictionary pairs. During the reconstruction pro- cess, simp e bicubic interpo ation approach is used for smooth patches w hi e edge and irregu ar structure patches are reconstructed from their corresponding dictionar y pairs using orthogona matching pursuit a gorithm . Experiment resu ts show that the propo sed a- gorithm signif icant y impro ves the visua qua ity of the edges and has f aster speed com