ABSTRACT
This paper presents a novel blind robust color image watermarking scheme using integer wavelet transform, bagging, and random subspace-KNN (RS-KNN) methods. Principal component analysis (PCA) reduces the number of features so that the training time can be reduced. The extraction of a watermark is considered here as a binary (0 or 1) classification problem. A binary watermark is implanted using some quantization approach. The suggested research is motivated by the excellent learning rate of the ensemble approach, which includes Bagging and the RS-KNN ML method. Two different forms of watermarks, namely, signature watermarks (the original watermark) and reference watermarks (produced at random), are utilized for embedding in the suggested research. Features are extracted from the colored watermarked image to form training and testing sets. It gives the imperceptibility of 37.83 dB, for Lena, 37.20 dB, for Peppers, 36.43 dB for Mandril, 37.26 dB for Jetplane, 34.72 dB for House, 36.58 dB for Lake, 36.98 dB for Car, and 36.56 dB for Zelda standard image. A thorough investigation has been conducted and provided both with and without PCA. The suggested watermarking method is examined in comparison to other machine learning-based image watermarking techniques and found to be durable against the majority of image attacks.
