计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2014年
20期
202-206
,共5页
双量化效应%盲检测%直方图%后验概率%连通区域
雙量化效應%盲檢測%直方圖%後驗概率%連通區域
쌍양화효응%맹검측%직방도%후험개솔%련통구역
double quantization effect%blind detection%histogram%posteriori probability%connected region
JPEG图像的双量化效应是检测和发现JPEG图像篡改的重要线索。针对现有检测算法大多数是基于DCT块效应而较少利用双量化效应的情况,提出了一种利用JPEG双量化效应的图像篡改盲检测新方案。该方案对比用DCT系数直方图计算的未篡改区域后验概率和用区间长度计算的未篡改区域后验概率之间的差异性,提取能有效区分篡改块和未篡改块的特征,计算每个DCT块的特征值,然后设置阈值,将特征值大于阈值的图像块判定为篡改块,最后通过选取连通区域,标定篡改区域。实验结果表明,与已有的类似方案相比,该方案能够较精确地检测和定位篡改区域,且对颜色、纹理丰富的图像具有明显优势。
JPEG圖像的雙量化效應是檢測和髮現JPEG圖像篡改的重要線索。針對現有檢測算法大多數是基于DCT塊效應而較少利用雙量化效應的情況,提齣瞭一種利用JPEG雙量化效應的圖像篡改盲檢測新方案。該方案對比用DCT繫數直方圖計算的未篡改區域後驗概率和用區間長度計算的未篡改區域後驗概率之間的差異性,提取能有效區分篡改塊和未篡改塊的特徵,計算每箇DCT塊的特徵值,然後設置閾值,將特徵值大于閾值的圖像塊判定為篡改塊,最後通過選取連通區域,標定篡改區域。實驗結果錶明,與已有的類似方案相比,該方案能夠較精確地檢測和定位篡改區域,且對顏色、紋理豐富的圖像具有明顯優勢。
JPEG도상적쌍양화효응시검측화발현JPEG도상찬개적중요선색。침대현유검측산법대다수시기우DCT괴효응이교소이용쌍양화효응적정황,제출료일충이용JPEG쌍양화효응적도상찬개맹검측신방안。해방안대비용DCT계수직방도계산적미찬개구역후험개솔화용구간장도계산적미찬개구역후험개솔지간적차이성,제취능유효구분찬개괴화미찬개괴적특정,계산매개DCT괴적특정치,연후설치역치,장특정치대우역치적도상괴판정위찬개괴,최후통과선취련통구역,표정찬개구역。실험결과표명,여이유적유사방안상비,해방안능구교정학지검측화정위찬개구역,차대안색、문리봉부적도상구유명현우세。
The double quantization effect in JPEG images is becoming important clue in detecting image forgeries. As many detection methods are based on the block artifacts of DCT coefficients, a new blind detection scheme which makes use of the double quantization effect is proposed. By comparing two posteriori probabilities of untampered region, which are computed by histogram of DCT coefficients and interval length respectively, a feature that can distinguish the tampered block and untampered block is extracted, and then the feature value of each DCT block is calculated. After setting a threshold, the tampered blocks whose feature values are larger than the threshold are obtained. Finally, the tampered region is located by labeling the connected components. Experimental results demonstrate that the proposed scheme can detect and localize the tampered areas more accurately with respect to previously proposed methods, especially for colorful or rich texture images.