多核学习纹理特征的无参考图像质量评价.pdfVIP

ISSN 1673．94l8 CODEN JKYTA8 E—mail：fcst@vip．163．corn JournalofFrontiersofComputerScienceandTechnology http：／／www．ceaj．org 1673--9418／2014／08(12)-·1517-·08 Tel：+86一l0 doi：10．3778~．issn．1673—9418．1405036 多核学习纹理特征的无参考图像质量评价 严大卫 ，桑庆兵 江南大学 物联网工程学院，江苏 无锡 214122 BlindImageQualityAssessmentwithTextureFeatureviaMultipleKernelLearning YANDawei，SANGQingbing SchoolofInternetofThingsEngineering，JiangnanUniversity,Wuxi，Jiangsu214122，China +Correspondingauthor：E-mail：229647702@qq．corn YANDawei，SANGQingbing．Blindimagequalityassessmentwithtexturefeatureviamultiplekernellearning． JournalofFrontiersofComputerScienceandTechnolog~2014，8(12)：1517-1524． Abstract：Becauseoftheirregularfeaturesextractedfrom imageswithvarioustypesofdistortion，thesinglekernel methodcannotgettheidealresult，thispaperpresentsanon-referenceimagequality evaluationmethodbasedon multiplekernellearningforvariouslearningtypesofdistortion．Firstlythispaperdoesaconversionofthegray scaleimagewiththestrucutretensorandthephasecongruencN thenextractsthesecondarystatisticalfeautresof graylevel—gradientCO—occurrencematrixrfom them，finallyinputsthesefeaturesintohierarchicalmultiplekernel learningmachinefortrainingandgetsthequaliyt score．Therandom experimentresultsonmultipleimagelibrary showthatthenewmethodresultsareconsistentwiththesubjectiveevaluationvalues，andhavebettergeneralization． Keywords：non—-referenceimagequaliyt assessment；multiplekernellearning；graylevel-·gradientCO·o·ccurrence matrix；structuretensor；phasecongruency 摘 要：由于多种失真类型灰度一共生矩阵特征的不规则性，单核方法无法取得理想结果，从而提出了一种基于 多核学习，针对多种失真类型的无参考图像质量评价方法。首先对灰度图像进行相位一致和结构张量变换 ， 得到相位一致 图和结构张量 图，然后分别对它们提取灰度．梯度共生矩阵二次统计特征 ，最后将提取的特征输 入到高效的分层多核学习机进行训练学习，预测得到图像的质量评分。多图像库多次随机实验结果表明，新