CAJ | 학술논문

为了解决数字图像多重复制粘贴篡改检测问题，克服广义2近邻(g2NN)算法对匹配特征点漏检的缺点，该文提出逆序广义2近邻(Rg2NN)算法。在计算匹配特征点时，该算法采用逆序方式计算特征点之间的匹配关系，可以更加准确地计算出所有与待检测特征点相匹配的特征点。实验证明，Rg2NN算法比g2NN算法计算出来的匹配特征点更加准确，提高了g2NN算法对多重复制粘贴篡改图像的检测能力，当图像中的一块区域被复制后在多处粘贴，或多块区域被复制粘贴时可以准确计算出复制粘贴区域。
위료해결수자도상다중복제점첩찬개검측문제，극복엄의2근린(g2NN)산법대필배특정점루검적결점，해문제출역서엄의2근린(Rg2NN)산법。재계산필배특정점시，해산법채용역서방식계산특정점지간적필배관계，가이경가준학지계산출소유여대검측특정점상필배적특정점。실험증명，Rg2NN산법비g2NN산법계산출래적필배특정점경가준학，제고료g2NN산법대다중복제점첩찬개도상적검측능력，당도상중적일괴구역피복제후재다처점첩，혹다괴구역피복제점첩시가이준학계산출복제점첩구역。
For the consideration of the multiple copy-move forgery detection of digital images, and to avoid missing the matching feature points when generalized 2 Nearest-Neighbor (g2NN) algorithm is applied, Reversed generalized 2 Nearest-Neighbor (Rg2NN) algorithm is proposed. Reverse order is used in feature points matching, so that all feature points that match with the detected point can be calculated accurately. The experiment results show that the matching feature points calculated by Rg2NN are more accurate than by g2NN, and the ability of g2NN in detecting multiple copy-move forgery is improved. When one patch in the image is copied and pasted multiple times or two or more patches are copied and pasted, the copy-move map can be localized precisely by the Rg2NN algorithm.