Digital image watermarking using fuzzy logic and naturalness preserving transform

• Authors: Ayman Mansour Tawfik Ahmed Elmousalami;Dwight D. Day
• Venue: Digital image watermarking using fuzzy logic and naturalness preserving transform
• Year: 2004

The research presented in this dissertation involves development of two image-watermarking approaches. The techniques are designed to protect conventional and satellite images from theft and illegal reproduction. During the course of this research, the scientific nature of remote sensing data has always been put forward as a reason to cautiously alter original data values. The first technique uses the Naturalness Preserving Transform (NPT) to encode identifying information into a host image. However, the second technique utilizes a Fuzzy Inference System (FIS) to extract the most robust portions from a host image. NPT was first introduced as an intermediate transform domain between the spatial representation of an image and its Hadamard transform. All watermarking techniques presented in the literature embed the watermark into host image either via the spatial or frequency representation of the image. The special characteristics of NPT in encoding information into images and retrieving missing portions from a corrupted image supported the idea of applying NPT to watermarking. The NPT-watermarking algorithm encodes identifying information (logo image) into host image via NPT transformation. An iterative algorithm is utilized to extract the hidden watermark. The algorithm involves enhancing the quality of the transformed image. Discrete cosine and Hartley matrices were used to modify the classical NPT to address the visual artifacts commonly introduced by the Hadamard transform. The watermark demonstrated robustness against a variety of image attacks: JPEG compression, filtering, cropping, and additive noise. The FIS watermarking technique introduces an approach to generating the watermarking mask based on extraction of the most robust regions of an image to host the watermark. The algorithm extracts the robust portions from the texture image based on the relationship between pixels within small blocks. Multiplying the output of the FIS by the watermark to generate the mask invisibly redistributes the identifying information. The watermark has been added to host images via the spatial and wavelet domains. This watermark also demonstrated robustness against JPEG compression and filtering attacks.