|
This paper describes a new approach for color image watermarking using Cycle Spinning based Sharp Frequency Localized Contourlet Transform and Principal Component Analysis. The approach starts with decomposition of images into various subbands using Contourlet Transform(CT) successively for all the color spaces of both host and watermark images. Then principal components of middle band(x bands) are considered for inserting operation. The ordinary contourlet transform suffers from lack of frequency localization. The localization being the most important criterion for watermarking, the conventional CT is not very suitable for watermarking. This problem of CT is over come by Sharp Frequency Localized Contourlet, but this lacks of translation invariance. Hence the cycle spinning based sharp frequency localized contourlet chosen for watermarking. Embedding at middle level sub bands(x band) preserves the curve nature of edges in the host image hence less disturbance is observed when host and watermark images are compared. This result in very good Peak Signal to Noise Ratio (PSNR) instead of directly adding of mid frequency components of watermark and host images the principal components are only added. Likewise the amount of payload to be added is reduced hence host images get very less distortion. Usage of principal components also helps in fruitful extraction of watermark information from host image hence gives good correlation between input watermark and extracted one. This technique has shown a very high robustness under various intentional and non intentional attacks.
|
