Many proposed digital image watermarking techniques are sensitive to geometric attacks, such as rotation, scaling, translation, or their composites. Geometric distortions, even by slight amounts, can inevitably damage the watermark and/or disable the capability of the watermark detector to reliably perform its function. In this chapter, the authors exploit the invariant image features to design geometric distortions-invariant watermarking system, and present two watermarking techniques. First technique utilizes the bounding box scale-invariant feature transform and discrete orthogonal Hahn moments to embed the watermark into the selective image patches, and the second technique uses only the Hahn moments to globally embed watermark into the whole image. First technique is non-blind and uses the original image during detection. While exhibiting excellent resistance against different geometric distortions, this technique also has fairly good resistance to image cropping like attacks. However, this technique exhibits a reduced data payload. The second technique is designed to be blind and the watermark is blindly extracted using the independent component analysis. For this technique an improved data payload is achieved but with a little compromise on resistance against cropping like attacks. The implementations are supported with thorough discussions and the experimental results prove and demonstrate the effectiveness of the proposed schemes against several kinds of geometric attacks.
GLAUCOMA IMAGE CLASSIFICATION USING DISCRETE ORTHOGONAL STOCKWELL TRANSFORM
