Independent component analysis in the blind watermarking of digital images Authors: J.J. Murillo-Fuentes Source: Science Direct Neurocomputing 70 (2007) 2881-2890 Speaker: Wei Chun, Lin

Outline Introduction ICA in image process ICA in watermarking of digital images A general ICA-based watermarking algorithm A practical RW algorithm Experimental results Conclusions

Introduction Image watermarking allows the protection of copyrighted data by ownership verification, preventing unauthorized distributions. Direct embedding in the spatial domain is desirable for low- complexity requirements. However, it may be easily attacked. Other techniques based on transformed domains are usually more robust. The basic idea of ICA RW is to project the image into independent components, the transform coefficients, to embed the watermark into them.

ICA in image process There are two common applications of ICA to image processing. 1. l linear mixtures of l images. 2. only one image is involved.

ICA in image process

ICA in image process

ICA in watermarking of digital images The cover (or original) image I and the watermark W are reshaped into vectors, xIt and xWt , and then the watermark is inserted into the cover image by mixing them.

A general ICA-based watermarking algorithm Embedding

A general ICA-based watermarking algorithm

A general ICA-based watermarking algorithm Detection

A general ICA-based watermarking algorithm

Experimental result

Experimental result

Conclusion The robustness of the method rests on two main ideas. 1. the ICA projection used is a key of the method as it is needed to completely remove the watermark. 2. the statistical and spectral characteristics of the independent components are close to that of the spread sequence.