Curvelet Based Watermarking Of Multispectral Images And Its Effect On Classification Accuracy

In this paper, a robust invisible watermarking algorithm for satellite imagery using Curvelet Transform is proposed. Haralick Co-occurrence texture features [1] are used to identify the area for watermarking in the host image. Host image is tiled into smaller non-overlapping blocks. Based on the Haralick texture feature chosen, blocks with high value of chosen texture feature were selected for embedding. Thus, multiple watermarks are embedded in any given image. There are some unstable Curvelet coefficients so a little change of the image will result in a big change of these coefficients. These unstable factors can influence the extracting of watermark. Hence selection of position of embedding in the transformed domain plays a very important role in robustness of the embedding process. This algorithm encourages use of edges and curves for embedding watermarks. The experimental results show that watermark using proposed algorithm is robust against common attacks like Brightness, Contrast, Saturation, Tint adjustments, Low pass Filtering, JPEG Compression attack, Gaussian Noise attack and Laplacian Filtering, Symmetrical and Asymmetrical Image Cropping, Geometric Attacks like scaling and Rotation. Use of SURF features [2] too ensures robustness to geometric attacks. The effect of watermarking on the classification accuracy is also studied.