| Input: A suspicious gray-scale image . |
| Output: All perceptual hash feature vectors for image |
| blocks in . |
| Step 1. Suspicious image is divided into |
| overlapping |
| blocks, denoted as , where , |
| , , and |
| . |
| Step 2. For each block |
| Step 3. The pixel mean of , denoted as , |
| is computed. |
| Step 4. DCT is applied to generate the coefficient |
| matrix for block , denoted as . |
| Step 5. The coefficient matrix is divided into |
| four sub-blocks, denoted as , , , |
| and , respectively. |
| Step 6. The mean of the first sub-block is |
| calculated, denoted as . |
| Step 7. The perceptual hashing matrix for each |
| sub-block is computed, denoted as |
| , where . |
| Step 8. Each perceptual hashing matrix is |
| converted into a decimal number, denoted |
| as , to represent feature value for block |
| , where . |
| Step 9. The feature vector of block is created, |
| denoted as , |
| according to its pixel mean and four |
| feature values. |
| Step 10. End For |
