Image based document authentication using DCT Source Pattern

Image based document authentication using DCT Source: Pattern Recognition Letters, Vol. 22, 2001, pp. 725 - 729 Author: Gaurav Garg, Pramod Kumar Sharma, Santanu Chaudhury Speaker: Chun-Sen Tseng Data: 2004/06/10 1

The step of proposed method n n Lowpass filter the original image so as to remove noise. Divide photo image to create m*n grid intersection point Around each grid to calculate DCT coefficients in different level neighbourhood area. Calculate the signature of each neighbourhood area. 2

Divide photo image to create m*n grid intersection point neighbourhood area with size 2 m*2 m Level 1：m=2 Level 2：m=3 Level 3：m=4 Level 4：m=5 3*3 grid intersection point 3

Around each grid to calculate DCT coefficients in different level neighbourhood area If level = 1 m=2 neighbourhood area with size 22*22 Calculate the DCT coefficient and quantized Add all the coefficients and take average to obtain P 1(i, j) 4

Example If level = 1 m=2 neighbourhood area with size 22*22 52 53 50 49 323 27 3 0 51 52 48 46 29 11 0 0 55 54 52 55 5 9 5 0 49 52 51 51 2 0 0 0 Pixel domain DCT domain P 1(0, 0) = ( 323/1 + 27/1 + 3/2 +…+ 0/4) /16 = 25. 825 5

Calculate the signature of each neighbourhood area 6

Example If p 1(0, 0) = 25. 875 , p 1(0, 1) = 32 , p 1(0, 2) = 15 , p 1(1, 0) = 12 , p 1(1, 1) = 22 , p 1(2, 1) = 45 , p 1(2, 0) = 5 , p 1(2, 1) = 52 , p 1(2, 2) = 17 , p 1(1, 1) < p 1(2, 1) s 1 =0 p 1(1, 1) > p 1(2, 2) ‧ ‧ p 1(1, 1) > p 1(0, 2) s 2 =1 S 1(1, 1) =10011010 s 8 =1 7

Photo image signature 8

Experiment result Original image After adding noise 9

Experiment result(cont. ) 10

Conclusion n This paper propose a link between photograph and personal identification documents. 11