Figure 3 ROC graphs for two tricks and two OCR engines
Related Figures (4)
Table 2 shows the results we obtain for the detection of two salting tricks, font color and font size. Table 2: Results for salting trick detection Figure 5: Detailed ROC graph for font size trick Table 3: Results for salting trick detection In this subsection we look into the individual distance metrics proposed in Section 4. Table 3 summarizes the classification results for both salting tricks and each individual distance metrics. We can observe that the TOC-feature is clearly the most useful one. For the font color trick the classifier based on this feature alone even outperforms the clas- sifier based on all features. On the other hand, the LENGTH and COMPLEXITY features are useful for the font color trick detection but not at all useful for the font size trick detection. Figure 4: Detailed ROC graph for font color trick
