电子科技大学学报
電子科技大學學報
전자과기대학학보
JOURNAL OF UNIVERSITY OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA
2013年
4期
603-608
,共6页
拷贝检测%几何失真%图像归一化%查准率%查全率
拷貝檢測%幾何失真%圖像歸一化%查準率%查全率
고패검측%궤하실진%도상귀일화%사준솔%사전솔
copy detection%geometric distortion%image normalization%precision%recall
提出一种基于不变矩的DCT域图像拷贝检测方法,选取图像亮度分量的整体DCT变换后的中低频系数的符号作为其DCT域特征,分别计算原始图像及其角度规范化后图像的DCT域特征,计算并比较查询图像与图像库中图像的两类特征的相关性,取较大的相关值来判断图像是否存在拷贝。图像的整体DCT中低频系数已被验证对于信号处理攻击鲁棒性高,且随着图像缩放攻击呈规律性变化,因而利用图像的整体DCT中低频系数构建的DCT域特征对于信号处理和图像缩放攻击鲁棒性高;利用图像的缩放不变矩估算出图像特征角度,再依据特征角度对图像进行角度规范化可消除旋转失真的影响,因而利用角度规范化后的图像的DCT域特征能抵抗旋转、以及旋转和等比例缩放的混合攻击。实验结果证明在抗旋转和包含旋转的混合攻击方面,该算法的检测效果明显优于已有算法。
提齣一種基于不變矩的DCT域圖像拷貝檢測方法,選取圖像亮度分量的整體DCT變換後的中低頻繫數的符號作為其DCT域特徵,分彆計算原始圖像及其角度規範化後圖像的DCT域特徵,計算併比較查詢圖像與圖像庫中圖像的兩類特徵的相關性,取較大的相關值來判斷圖像是否存在拷貝。圖像的整體DCT中低頻繫數已被驗證對于信號處理攻擊魯棒性高,且隨著圖像縮放攻擊呈規律性變化,因而利用圖像的整體DCT中低頻繫數構建的DCT域特徵對于信號處理和圖像縮放攻擊魯棒性高;利用圖像的縮放不變矩估算齣圖像特徵角度,再依據特徵角度對圖像進行角度規範化可消除鏇轉失真的影響,因而利用角度規範化後的圖像的DCT域特徵能牴抗鏇轉、以及鏇轉和等比例縮放的混閤攻擊。實驗結果證明在抗鏇轉和包含鏇轉的混閤攻擊方麵,該算法的檢測效果明顯優于已有算法。
제출일충기우불변구적DCT역도상고패검측방법,선취도상량도분량적정체DCT변환후적중저빈계수적부호작위기DCT역특정,분별계산원시도상급기각도규범화후도상적DCT역특정,계산병비교사순도상여도상고중도상적량류특정적상관성,취교대적상관치래판단도상시부존재고패。도상적정체DCT중저빈계수이피험증대우신호처리공격로봉성고,차수착도상축방공격정규률성변화,인이이용도상적정체DCT중저빈계수구건적DCT역특정대우신호처리화도상축방공격로봉성고;이용도상적축방불변구고산출도상특정각도,재의거특정각도대도상진행각도규범화가소제선전실진적영향,인이이용각도규범화후적도상적DCT역특정능저항선전、이급선전화등비례축방적혼합공격。실험결과증명재항선전화포함선전적혼합공격방면,해산법적검측효과명현우우이유산법。
In this paper, an invariant moment based robust image copy detection algorithm in discrete cosine transformation (DCT) domain is proposed. Low and middle frequency coefficients which are calculated from full DCT transformation on luminance component of image are used as the image eigenvector in DCT domain. The query image and its angle normalized image eigenvector in DCT domain are computed, then correlation values of two eigenvectors of the query image and those of test image are respectively computed and the larger one is used to judge whether the query image is a copy or not. Full DCT coefficients are verified to have high robustness against common signal attacks and regularly vary with scaling attacks, thus the DCT eigenvector constructed by middle and low frequency full DCT coefficients can resist signal attacks and scaling distortion. The characteristic angle estimated by the image scale invariant moment is used to normalize the image to counteract the rotation distortion, so the angle normalized image can resist rotation distortion. The experimental results show that the algorithm is superior to Ling’s methods in terms of rotation distortion.