How to Reconstruct the History of a Digital Image, and of Its Alterations

Image Forgery Detection Based on Noise Inspection: Analysis and Refinement of the Noisesniffer Method

“…We used the Trace database [3] to evaluate the method. The Trace database is made of 1000 images taken from the Raise dataset [9].…”

“…We observe that the size of the blocks w highly impacts the performance of the method, being 3 the optimal value among the tested ones. 2: Scores obtained on the endomasks noise level dataset from the Trace database [3] when setting w to 3, 5 and 8, while keeping the rest of the parameters to their optimal value. Figure 1 shows an example of the results obtained when setting w to 3, 5 and 8.…”

“…Indeed, when setting w to 3 the M CC score is 0.9322, the F 1 score 0.9444, and the IoU score 0.8947 while the corresponding scores when setting w to 5 and 8 are 0.9275, 0.9404, 0.8875 and 0.8730, 0.8965, 0.81239, respectively. Figure 1: Results obtained on image r0f22c43ct from the Trace dataset [3] when setting w to 3, 5, and 8, while keeping the rest of the parameters to their optimal value. The distribution images show the blocks in L painted in white and, on top, those in V painted in red.…”

“…n MCC F1 IoU 0.1 0.3572 0.3716 0.2723 0.05 0.3372 0.3382 0.2437 0.01 0.2020 0.1864 0.1295 Table 4: Scores obtained on endomasks noise level dataset from the Trace database [3] when setting n to 0.1, 0.05, and 0.01, while keeping the rest of the parameters to their optimal value.…”

“…m MCC F1 IoU 0.5 0.3572 0.3716 0.2723 0.4 0.3540 0.3716 0.2723 0.3 0.3498 0.3600 0.2607 0.2 0.3388 0.3437 0.2466 0.1 0.3166 0.3155 0.2234 Table 5: Scores obtained on the endomasks noise level dataset from the Trace database [3] when setting m to 0.5, 0.4, 0.3, 0.2 and 0.1, while keeping the rest of the parameters to their optimal value.…”