150 likes | 276 Views
Feature Space Based Watermarking in Multi-Images. Xin Zhou. Outlines. Introduction Feature Space Based Watermarking Simulation Results Conclusion. Introduction. Goal: Implement the watermarking in a set of images or video based on eigen-decomposition or SVD (Singular Value Decomposition).
E N D
Outlines • Introduction • Feature Space Based Watermarking • Simulation Results • Conclusion
Introduction • Goal: • Implement the watermarking in a set of images or video based on eigen-decomposition or SVD (Singular Value Decomposition)
Introduction(2) • Where to embed bits? • Spatial Domain: Directly process original values of the host image according to the watermark • Frequency Domain: Transfer the host image into another domain and change the corresponding coefficients according to the watermark • Is it possible to embed bits in the feature domain?
Introduction(3) • Other’s Work • Using SVD (Singular Value Decomposition) to embed bits in one image • Using ICA (Independent Component Analysis) to detect the watermark • What I want to do • Find a method to use eigen-decomposition or SVD to embed watermark in multi-images or video • It will be more robust to embed bits in each images • More difficult for others to estimate the watermark
Feature Based Watermarking • Eigen-Decomposition and SVD • Embed one bit • Detection • Embed watermark in a set of images
Eigen-Decomposition and SVD • mxn matrix A, of rank r, can be expressed as the product: A = U * S * VT • U is mxr term matrix • S is rxr diagonal matrix • V is rxn document matrix • If A is nxn matrix, r=n, we have A *U= U * S
Embedding One Bit (1) • Assume the host image is a mxn matrix • Perform the SVD to get S matrix • Embed one bit in the S matrix according to • where {si}: original coefficients {si’}: marked coefficients {b}: the bit to be embedded which is 0 or 1 k : watermark strength, adjusted by the just-noticeable- difference (JND) standard
Embedding One Bit(2) After Embedding: A’ = U * S’ * VT where S’ is the watermarked singular matrix A’ is the corresponding watermarked image
Detection • Assume we get the watermarked image A’. • Perform eigen-decomposition or SVD to get the S’ • Compare S and S’, we can get the watermark
Embedding in Multi-Images • Method I: • Generate a pseudo random codebook • Decide which bit should be embedded to which image based on the codebook • For a specific image, use the previous method to embed bits • Method II: • Use QIM-like method to decide which bit should be embedded to which image.
Conclusion • Using eigen-decomposition or SVD to embed watermark into multi-images. • Implemented basic functions of the proposed method • Need to do more tests under various attacks