standard:
1) imperceptibility 2)robustness 3)capacity 4)security 5)false -arlarm rate
algorithms:
1) patchwork: divide image into two collections with equal size
if w=1, add 1 to the first colletion,while decrease 1 to the other
if w=0,decrease 1 to the first collection,while add 1 to the other/
statistic: compare the two collections ,if the first is more than the other, represents w=1,which is inserted.
questions: how are the two collections is divided? how to identify which is which?
2)Generate Guassian: make noise N with the same dimension to image I
then: if w=1, I'= I+N;
w=0,I'=I-N;
c=I' * N
statistic: check out the c is more than 0;
question: what's the dimension means? why c must be the multiply of i' and N?
3)spread spectrum: 1) perform DCT to image
2)Generate Guassian-like noise vextor x;
3)select 1000 max DCT coefficients;
then vi=vi(1+aXi)
SUM( X,X*)
question: vextor?
4)LSB: Generate m sequence
Compress original image to 128 color using adaptive histogram algorithm
Insert m-sequence to LSB of image
Detection: correlation of LSB to m-sequence
5) predictive coding:
Generate a difference table with key
Modify pixel with smallest change so that the difference of two adjacent corresponds to current inserted bit.
6)QIM:
7)Salient point:
8) Chaos based
9) RST -invariant