Feature Selection of the Rich Model Based on the Correlation of Feature Components
Algorithm 1
Selection algorithm of correlation coefficient threshold T.
(i)
Input: steganalysis feature of cover images , steganalysis feature of stegoimages .
(ii)
Output: the proper correlation coefficient threshold .
(iii)
Step 1: calculate the initial detection accuracy. According to , obtain 9 values ,, ……, with an interval of 0.01, and calculate their corresponding detection accuracy ,, ……, and the detection accuracy of original feature .
(iv)
Step 2: compare detection performance. Compare with in the order of . When , stop subsequent comparisons and assign and to and , respectively.
(v)
Step 3: reduce the selection range of T. Let , calculate its corresponding detection accuracy , and compare with . If , then assign to ; otherwise, assign to .
(vi)
Step 4: repeat Step 3 until the significant digits of reach 5, and stop the operation and regard as the optimal truncation threshold .