A Multiobjective Evolutionary Algorithm for Hyperspectral Image Watermarking
With the increasing availability of internet access to remote sensing imagery, the concern with image authentication and ownership issues is growing in the remote sensing community. Watermarking techniques help to solve the problems raised by this issue.
- PDF / 905,088 Bytes
- 16 Pages / 430 x 660 pts Page_size
- 60 Downloads / 227 Views
roduction The hyperspectral sensor performs a fine sampling of the surface radiance in the visible and near infrared wavelength spectrum. Therefore each image pixel may be interpreted as a high dimensional vector. We are interested in the watermarking of hyperspectral images because all the new remote sensor are designed to be hyperspectral. The fact that Internet is is becoming the primary mean of communication and transport of these images, may raise authentication and ownership issues in the near future. Watermarking is a technique for image authorship and content protection [21, 1, 15, 16, 20, 22, 13, 23]. Semi-fragile watermarking [12, 24] tries to ensure the image integrity, by means of an embedded watermark which can be recovered without modification if the image has not been manipulated. However, it is desirable that the watermark recovery is robust to operations like filtering, smoothing and lossy compression [19] which are very common while distributing
The Spanish Ministerio de Educacion y Ciencia supports this work through grant DPI2006-15346-C03-03 and VIMS-2003-20088-c04-04.
M. Gra˜ na and R.J. Duro (Eds.): Comput. Intel. for Remote Sensing, SCI 133, pp. 63–78, 2008. c Springer-Verlag Berlin Heidelberg 2008 springerlink.com
64
D. Sal and M. Gra˜ na
images through communication networks. For instance, the JPEG lossy compression first standard deletes the image discrete cosine transform (DCT) high frequency coefficients. The JPEG 2000 standard works on the image discrete wavelet transform (DWT) coefficients, also removing high frequency ones as needed to attain the desired compression ratio. Embedding the watermark image in the image transform coefficients is the usual and most convenient approach when trying to obtain perceptually invisible watermarks. We have focused on the DCT transform for several reasons. First it is a real valued transform, so we do not need to deal with complex numbers. Second, the transform domain is continuously evolving from low to high spatial frequencies, unlike DWT which has a complex hierarchical structure in the transform domain. The definition of the fitness functions below benefits from this domain continuity. It is possible to assume some conclusions about the watermark robustness dependence on its placement. Besides being robust, we want the watermarked image must be as perceptually indistinguishable from the original one as possible, that is, the watermarking process must introduce the minimum possible visual distortion in the image. These two requirements (robustness against filtering and minimal distortion) are the contradicting objectives of our work. The trivial watermarking approach consists in the addition or substitution of the watermark image over the high frequency image transform coefficients. That way, the distortion is perceptually minimal, because the watermark is embedded in the noisy components of the image. However, this approach is not robust against smoothing and lossy compression. The robustness can be enhanced placing the watermark in other regions of the
Data Loading...
